Tag Archives: Computer Vision

3D Scene Understanding with TensorFlow 3D

The growing ubiquity of 3D sensors (e.g., Lidar, depth sensing cameras and radar) over the last few years has created a need for scene understanding technology that can process the data these devices capture. Such technology can enable machine learning (ML) systems that use these sensors, like autonomous cars and robots, to navigate and operate in the real world, and can create an improved augmented reality experience on mobile devices. The field of computer vision has recently begun making good progress in 3D scene understanding, including models for mobile 3D object detection, transparent object detection, and more, but entry to the field can be challenging due to the limited availability tools and resources that can be applied to 3D data.

In order to further improve 3D scene understanding and reduce barriers to entry for interested researchers, we are releasing TensorFlow 3D (TF 3D), a highly modular and efficient library that is designed to bring 3D deep learning capabilities into TensorFlow. TF 3D provides a set of popular operations, loss functions, data processing tools, models and metrics that enables the broader research community to develop, train and deploy state-of-the-art 3D scene understanding models.

TF 3D contains training and evaluation pipelines for state-of-the-art 3D semantic segmentation, 3D object detection and 3D instance segmentation, with support for distributed training. It also enables other potential applications like 3D object shape prediction, point cloud registration and point cloud densification. In addition, it offers a unified dataset specification and configuration for training and evaluation of the standard 3D scene understanding datasets. It currently supports the Waymo Open, ScanNet, and Rio datasets. However, users can freely convert other popular datasets, such as NuScenes and Kitti, into a similar format and use them in the pre-existing or custom created pipelines, and can leverage TF 3D for a wide variety of 3D deep learning research and applications, from quickly prototyping and trying new ideas to deploying a real-time inference system.

An example output of the 3D object detection model in TF 3D on a frame from Waymo Open Dataset is shown on the left. An example output of the 3D instance segmentation model on a scene from ScanNet dataset is shown on the right.

Here, we will present the efficient and configurable sparse convolutional backbone that is provided in TF 3D, which is the key to achieving state-of-the-art results on various 3D scene understanding tasks. Furthermore, we will go over each of the three pipelines that TF 3D currently supports: 3D semantic segmentation, 3D object detection and 3D instance segmentation.

3D Sparse Convolutional Network
The 3D data captured by sensors often consists of a scene that contains a set of objects of interest (e.g. cars, pedestrians, etc.) surrounded mostly by open space, which is of limited (or no) interest. As such, 3D data is inherently sparse. In such an environment, standard implementation of convolutions would be computationally intensive and consume a large amount of memory. So, in TF 3D we use submanifold sparse convolution and pooling operations, which are designed to process 3D sparse data more efficiently. Sparse convolutional models are core to the state-of-the-art methods applied in most outdoor self-driving (e.g. Waymo, NuScenes) and indoor benchmarks (e.g. ScanNet).

We also use various CUDA techniques to speed up the computation (e.g., hashing, partitioning / caching the filter in shared memory, and using bit operations). Experiments on the Waymo Open dataset shows that this implementation is around 20x faster than a well-designed implementation with pre-existing TensorFlow operations.

TF 3D then uses the 3D submanifold sparse U-Net architecture to extract a feature for each voxel. The U-Net architecture has proven to be effective by letting the network extract both coarse and fine features and combining them to make the predictions. The U-Net network consists of three modules, an encoder, a bottleneck, and a decoder, each of which consists of a number of sparse convolution blocks with possible pooling or un-pooling operations.

A 3D sparse voxel U-Net architecture. Note that a horizontal arrow takes in the voxel features and applies a submanifold sparse convolution to it. An arrow that is moving down performs a submanifold sparse pooling. An arrow that is moving up will gather back the pooled features, concatenate them with the features coming from the horizontal arrow, and perform a submanifold sparse convolution on the concatenated features.

The sparse convolutional network described above is the backbone for the 3D scene understanding pipelines that are offered in TF 3D. Each of the models described below uses this backbone network to extract features for the sparse voxels, and then adds one or multiple additional prediction heads to infer the task of interest. The user can configure the U-Net network by changing the number of encoder / decoder layers and the number of convolutions in each layer, and by modifying the convolution filter sizes, which enables a wide range of speed / accuracy tradeoffs to be explored through the different backbone configurations

3D Semantic Segmentation
The 3D semantic segmentation model has only one output head for predicting the per-voxel semantic scores, which are mapped back to points to predict a semantic label per point.

3D semantic segmentation of an indoor scene from ScanNet dataset.

3D Instance Segmentation
In 3D instance segmentation, in addition to predicting semantics, the goal is to group the voxels that belong to the same object together. The 3D instance segmentation algorithm used in TF 3D is based on our previous work on 2D image segmentation using deep metric learning. The model predicts a per-voxel instance embedding vector as well as a semantic score for each voxel. The instance embedding vectors map the voxels to an embedding space where voxels that correspond to the same object instance are close together, while those that correspond to different objects are far apart. In this case, the input is a point cloud instead of an image, and it uses a 3D sparse network instead of a 2D image network. At inference time, a greedy algorithm picks one instance seed at a time, and uses the distance between the voxel embeddings to group them into segments.

3D Object Detection
The 3D object detection model predicts per-voxel size, center, and rotation matrices and the object semantic scores. At inference time, a box proposal mechanism is used to reduce the hundreds of thousands of per-voxel box predictions into a few accurate box proposals, and then at training time, box prediction and classification losses are applied to per-voxel predictions. We apply a Huber loss on the distance between predicted and the ground-truth box corners. Since the function that estimates the box corners from its size, center and rotation matrix is differentiable, the loss will automatically propagate back to those predicted object properties. We use a dynamic box classification loss that classifies a box that strongly overlaps with the ground-truth as positive and classifies the non-overlapping boxes as negative.

Our 3D object detection results on ScanNet dataset.

In our recent paper, “DOPS: Learning to Detect 3D Objects and Predict their 3D Shapes”, we describe in detail the single-stage weakly supervised learning algorithm used for object detection in TF 3D. In addition, in a follow up work, we extended the 3D object detection model to leverage temporal information by proposing a sparse LSTM-based multi-frame model. We go on to show that this temporal model outperforms the frame-by-frame approach by 7.5% in the Waymo Open dataset.

The 3D object detection and shape prediction model introduced in the DOPS paper. A 3D sparse U-Net is used to extract a feature vector for each voxel. The object detection module uses these features to propose 3D boxes and semantic scores. At the same time, the other branch of the network predicts a shape embedding that is used to output a mesh for each object.

Ready to Get Started?
We’ve certainly found this codebase to be useful for our 3D computer vision projects, and we hope that you will as well. Contributions to the codebase are welcome and please stay tuned for our own further updates to the framework. To get started please visit our github repository.

Acknowledgements
The release of the TensorFlow 3D codebase and model has been the result of widespread collaboration among Google researchers with feedback and testing from product groups. In particular we want to highlight the core contributions by Alireza Fathi and Rui Huang (work performed while at Google), with special additional thanks to Guangda Lai, Abhijit Kundu, Pei Sun, Thomas Funkhouser, David Ross, Caroline Pantofaru, Johanna Wald, Angela Dai and Matthias Niessner.

Source: Google AI Blog


3D Scene Understanding with TensorFlow 3D

The growing ubiquity of 3D sensors (e.g., Lidar, depth sensing cameras and radar) over the last few years has created a need for scene understanding technology that can process the data these devices capture. Such technology can enable machine learning (ML) systems that use these sensors, like autonomous cars and robots, to navigate and operate in the real world, and can create an improved augmented reality experience on mobile devices. The field of computer vision has recently begun making good progress in 3D scene understanding, including models for mobile 3D object detection, transparent object detection, and more, but entry to the field can be challenging due to the limited availability tools and resources that can be applied to 3D data.

In order to further improve 3D scene understanding and reduce barriers to entry for interested researchers, we are releasing TensorFlow 3D (TF 3D), a highly modular and efficient library that is designed to bring 3D deep learning capabilities into TensorFlow. TF 3D provides a set of popular operations, loss functions, data processing tools, models and metrics that enables the broader research community to develop, train and deploy state-of-the-art 3D scene understanding models.

TF 3D contains training and evaluation pipelines for state-of-the-art 3D semantic segmentation, 3D object detection and 3D instance segmentation, with support for distributed training. It also enables other potential applications like 3D object shape prediction, point cloud registration and point cloud densification. In addition, it offers a unified dataset specification and configuration for training and evaluation of the standard 3D scene understanding datasets. It currently supports the Waymo Open, ScanNet, and Rio datasets. However, users can freely convert other popular datasets, such as NuScenes and Kitti, into a similar format and use them in the pre-existing or custom created pipelines, and can leverage TF 3D for a wide variety of 3D deep learning research and applications, from quickly prototyping and trying new ideas to deploying a real-time inference system.

An example output of the 3D object detection model in TF 3D on a frame from Waymo Open Dataset is shown on the left. An example output of the 3D instance segmentation model on a scene from ScanNet dataset is shown on the right.

Here, we will present the efficient and configurable sparse convolutional backbone that is provided in TF 3D, which is the key to achieving state-of-the-art results on various 3D scene understanding tasks. Furthermore, we will go over each of the three pipelines that TF 3D currently supports: 3D semantic segmentation, 3D object detection and 3D instance segmentation.

3D Sparse Convolutional Network
The 3D data captured by sensors often consists of a scene that contains a set of objects of interest (e.g. cars, pedestrians, etc.) surrounded mostly by open space, which is of limited (or no) interest. As such, 3D data is inherently sparse. In such an environment, standard implementation of convolutions would be computationally intensive and consume a large amount of memory. So, in TF 3D we use submanifold sparse convolution and pooling operations, which are designed to process 3D sparse data more efficiently. Sparse convolutional models are core to the state-of-the-art methods applied in most outdoor self-driving (e.g. Waymo, NuScenes) and indoor benchmarks (e.g. ScanNet).

We also use various CUDA techniques to speed up the computation (e.g., hashing, partitioning / caching the filter in shared memory, and using bit operations). Experiments on the Waymo Open dataset shows that this implementation is around 20x faster than a well-designed implementation with pre-existing TensorFlow operations.

TF 3D then uses the 3D submanifold sparse U-Net architecture to extract a feature for each voxel. The U-Net architecture has proven to be effective by letting the network extract both coarse and fine features and combining them to make the predictions. The U-Net network consists of three modules, an encoder, a bottleneck, and a decoder, each of which consists of a number of sparse convolution blocks with possible pooling or un-pooling operations.

A 3D sparse voxel U-Net architecture. Note that a horizontal arrow takes in the voxel features and applies a submanifold sparse convolution to it. An arrow that is moving down performs a submanifold sparse pooling. An arrow that is moving up will gather back the pooled features, concatenate them with the features coming from the horizontal arrow, and perform a submanifold sparse convolution on the concatenated features.

The sparse convolutional network described above is the backbone for the 3D scene understanding pipelines that are offered in TF 3D. Each of the models described below uses this backbone network to extract features for the sparse voxels, and then adds one or multiple additional prediction heads to infer the task of interest. The user can configure the U-Net network by changing the number of encoder / decoder layers and the number of convolutions in each layer, and by modifying the convolution filter sizes, which enables a wide range of speed / accuracy tradeoffs to be explored through the different backbone configurations

3D Semantic Segmentation
The 3D semantic segmentation model has only one output head for predicting the per-voxel semantic scores, which are mapped back to points to predict a semantic label per point.

3D semantic segmentation of an indoor scene from ScanNet dataset.

3D Instance Segmentation
In 3D instance segmentation, in addition to predicting semantics, the goal is to group the voxels that belong to the same object together. The 3D instance segmentation algorithm used in TF 3D is based on our previous work on 2D image segmentation using deep metric learning. The model predicts a per-voxel instance embedding vector as well as a semantic score for each voxel. The instance embedding vectors map the voxels to an embedding space where voxels that correspond to the same object instance are close together, while those that correspond to different objects are far apart. In this case, the input is a point cloud instead of an image, and it uses a 3D sparse network instead of a 2D image network. At inference time, a greedy algorithm picks one instance seed at a time, and uses the distance between the voxel embeddings to group them into segments.

3D Object Detection
The 3D object detection model predicts per-voxel size, center, and rotation matrices and the object semantic scores. At inference time, a box proposal mechanism is used to reduce the hundreds of thousands of per-voxel box predictions into a few accurate box proposals, and then at training time, box prediction and classification losses are applied to per-voxel predictions. We apply a Huber loss on the distance between predicted and the ground-truth box corners. Since the function that estimates the box corners from its size, center and rotation matrix is differentiable, the loss will automatically propagate back to those predicted object properties. We use a dynamic box classification loss that classifies a box that strongly overlaps with the ground-truth as positive and classifies the non-overlapping boxes as negative.

Our 3D object detection results on ScanNet dataset.

In our recent paper, “DOPS: Learning to Detect 3D Objects and Predict their 3D Shapes”, we describe in detail the single-stage weakly supervised learning algorithm used for object detection in TF 3D. In addition, in a follow up work, we extended the 3D object detection model to leverage temporal information by proposing a sparse LSTM-based multi-frame model. We go on to show that this temporal model outperforms the frame-by-frame approach by 7.5% in the Waymo Open dataset.

The 3D object detection and shape prediction model introduced in the DOPS paper. A 3D sparse U-Net is used to extract a feature vector for each voxel. The object detection module uses these features to propose 3D boxes and semantic scores. At the same time, the other branch of the network predicts a shape embedding that is used to output a mesh for each object.

Ready to Get Started?
We’ve certainly found this codebase to be useful for our 3D computer vision projects, and we hope that you will as well. Contributions to the codebase are welcome and please stay tuned for our own further updates to the framework. To get started please visit our github repository.

Acknowledgements
The release of the TensorFlow 3D codebase and model has been the result of widespread collaboration among Google researchers with feedback and testing from product groups. In particular we want to highlight the core contributions by Alireza Fathi and Rui Huang (work performed while at Google), with special additional thanks to Guangda Lai, Abhijit Kundu, Pei Sun, Thomas Funkhouser, David Ross, Caroline Pantofaru, Johanna Wald, Angela Dai and Matthias Niessner.

Source: Google AI Blog


Improving Mobile App Accessibility with Icon Detection

Voice Access enables users to control their Android device hands free, using only verbal commands. In order to function properly, it needs on-screen user interface (UI) elements to have reliable accessibility labels, which are provided to the operating system’s accessibility services via the accessibility tree. Unfortunately, in many apps, adequate labels aren’t always available for UI elements, e.g. images and icons, reducing the usability of Voice Access.

The Voice Access app extracts elements from the view hierarchy to localize and annotate various UI elements. It can provide a precise description for elements that have an explicit content description. On the other hand, the absence of content description can result in many unrecognized elements undermining the ability of Voice Access to function with some apps.

Addressing this challenge requires a system that can automatically detect icons using only the pixel values displayed on the screen, regardless of whether icons have been given suitable accessibility labels. What little research exists on this topic typically uses classifiers, sometimes combined with language models to infer classes and attributes from UI elements. However, these classifiers still rely on the accessibility tree to obtain bounding boxes for UI elements, and fail when appropriate labels do not exist.

Here, we describe IconNet, a vision-based object detection model that can automatically detect icons on the screen in a manner that is agnostic to the underlying structure of the app being used, launched as part of the latest version of Voice Access. IconNet can detect 31 different icon types (to be extended to more than 70 types soon) based on UI screenshots alone. IconNet is optimized to run on-device for mobile environments, with a compact size and fast inference time to enable a seamless user experience. The current IconNet model achieves a mean average precision (mAP) of 94.2% running at 9 FPS on a Pixel 3A.

Voice Access 5.0: the icons detected by IconNet can now be referred to by their names.

Detecting Icons in Screenshots
From a technical perspective, the problem of detecting icons on app screens is similar to classical object detection, in that individual elements are labelled by the model with their locations and sizes. But, in other ways, it’s quite different. Icons are typically small objects, with relatively basic geometric shapes and a limited range of colors, and app screens widely differ from natural images in that they are more structured and geometrical.

A significant challenge in the development of an on-device UI element detector for Voice Access is that it must be able to run on a wide variety of phones with a range of performance performance capabilities, while preserving the user’s privacy. For a fast user experience, a lightweight model with low inference latency is needed. Because Voice Access needs to use the labels in response to an utterance from a user (e.g., “tap camera”, or “show labels”) inference time needs to be short (<150 ms on a Pixel 3A) with a model size less than 10 MB.

IconNet
IconNet is based on the novel CenterNet architecture, which extracts features from input images and then predicts appropriate bounding box centers and sizes (in the form of heatmaps). CenterNet is particularly suited here because UI elements consist of simple, symmetric geometric shapes, making it easier to identify their centers than for natural images. The total loss used is a combination of a standard L1 loss for the icon sizes and a modified CornerNet Focal loss for the center predictions, the latter of which addresses icon class imbalances between commonly occurring icons (e.g., arrow backward, menu, more, and star) and underrepresented icons (end call, delete, launch apps, etc.)..

After experimenting with several backbones (MobileNet, ResNet, UNet, etc), we selected the most promising server-side architecture — Hourglass — as a starting point for designing a backbone tailored for icon and UI element detection. While this architecture is perfectly suitable for server side models, vanilla Hourglass backbones are not an option for a model that will run on a mobile device, due to their large size and slow inference time. We restricted our on-device network design to a single stack, and drastically reduced the width of the backbone. Furthermore, as the detection of icons relies on more local features (compared to real objects), we could further reduce the depth of the backbone without adversely affecting the performance. Ablation studies convinced us of the importance of skip connections and high resolution features. For example, trimming skip connections in the final layer reduced the mAP by 1.5%, and removing such connections from both the final and penultimate layers resulted in a decline of 3.5% mAP.

IconNet analyzes the pixels of the screen and identifies the centers of icons by generating heatmaps, which provide precise information about the position and type of the different types of icons present on the screen. This enables Voice Access users to refer to these elements by their name (e.g., “Tap ‘menu”).

Model Improvements
Once the backbone architecture was selected, we used neural architecture search (NAS) to explore variations on the network architecture and uncover an optimal set of training and model parameters that would balance model performance (mAP) with latency (FLOPs). Additionally, we used Fine-Grained Stochastic Architecture Search (FiGS) to further refine the backbone design. FiGS is a differentiable architecture search technique that uncovers sparse structures by pruning a candidate architecture and discarding unnecessary connections. This technique allowed us to reduce the model size by 20% without any loss in performance, and by 50% with only a minor drop of 0.3% in mAP.

Improving the quality of the training dataset also played an important role in boosting the model performance. We collected and labeled more than 700K screenshots, and in the process, we streamlined data collection by using heuristics and auxiliary models to identify rarer icons. We also took advantage of data augmentation techniques by enriching existing screenshots with infrequent icons.

To improve the inference time, we modified our model to run using Neural Networks API (NNAPI) on a variety of Qualcomm DSPs available on many mobile phones. For this we converted the model to use 8-bit integer quantization which gives the additional benefit of model size reduction. After some experimentation, we used quantization aware training to quantize the model, while matching the performance of a server-side floating point model. The quantized model results in a 6x speed-up (700ms vs 110ms) and 50% size reduction while losing only ~0.5% mAP compared to the unquantized model.

Results
We use traditional object detection metrics (e.g., mAP) to measure model performance. In addition, to better capture the use case of voice controlled user actions, we define a modified version of a false positive (FP) detection, where we penalize more incorrect detections for icon classes that are present on the screen. For comparing detections with ground truth, we use the center in region of interest (CIROI), another metric we developed for this work, which returns in a positive match when the center of the detected bounding box lies inside the ground truth bounding box. This better captures the Voice Access mode of operation, where actions are performed by tapping anywhere in the region of the UI element of interest.

We compared the IconNet model with various other mobile compatible object detectors, including MobileNetEdgeTPU and SSD MobileNet v2. Experiments showed that for a fixed latency, IconNet outperformed the other models in terms of [email protected] on our internal evaluation set.

Model    [email protected]
IconNet (Hourglass)    96%
IconNet (HRNet)    89%
MobilenetEdgeTPU (AutoML)    91%
SSD Mobilenet v2    88%

The performance advantage of IconNet persists when considering quantized models and models for a fixed latency budget.

Models (Quantized)    [email protected]    Model size    Latency*
IconNet (Currently deployed)    94.20%    8.5 MB    107 ms
IconNet (XS)    92.80%    2.3 MB    102 ms
IconNet (S)    91.70%    4.4 MB    45 ms
MobilenetEdgeTPU (AutoML)    88.90%    7.8 MB    26 ms
*Measured on Pixel 3A.

Conclusion and Future Work
We are constantly working on improving IconNet. Among other things, we are interested in increasing the range of elements supported by IconNet to include any generic UI element, such as images, text, or buttons. We also plan to extend IconNet to differentiate between similar looking icons by identifying their functionality. On the application side, we are hoping to increase the number of apps with valid content descriptions by augmenting developer tools to suggest content descriptions for different UI elements when building applications.

Acknowledgements
This project is the result of joint work with Maria Wang, Tautvydas Misiūnas, Lijuan Liu, Ying Xu, Nevan Wichers, Xiaoxue Zang, Gabriel Schubiner, Abhinav Rastogi, Jindong (JD) Chen, Abhanshu Sharma, Pranav Khaitan, Matt Sharifi and Blaise Aguera y Arcas. We sincerely thank our collaborators Robert Berry, Folawiyo Campbell, Shraman Ray Chaudhuri, Nghi Doan, Elad Eban, Marybeth Fair, Alec Go, Sahil Goel, Tom Hume, Cassandra Luongo, Yair Movshovitz-Attias, James Stout, Gabriel Taubman and Anton Vayvod. We are very grateful to Tom Small for assisting us in preparing the post.

Source: Google AI Blog


Improving Mobile App Accessibility with Icon Detection

Voice Access enables users to control their Android device hands free, using only verbal commands. In order to function properly, it needs on-screen user interface (UI) elements to have reliable accessibility labels, which are provided to the operating system’s accessibility services via the accessibility tree. Unfortunately, in many apps, adequate labels aren’t always available for UI elements, e.g. images and icons, reducing the usability of Voice Access.

The Voice Access app extracts elements from the view hierarchy to localize and annotate various UI elements. It can provide a precise description for elements that have an explicit content description. On the other hand, the absence of content description can result in many unrecognized elements undermining the ability of Voice Access to function with some apps.

Addressing this challenge requires a system that can automatically detect icons using only the pixel values displayed on the screen, regardless of whether icons have been given suitable accessibility labels. What little research exists on this topic typically uses classifiers, sometimes combined with language models to infer classes and attributes from UI elements. However, these classifiers still rely on the accessibility tree to obtain bounding boxes for UI elements, and fail when appropriate labels do not exist.

Here, we describe IconNet, a vision-based object detection model that can automatically detect icons on the screen in a manner that is agnostic to the underlying structure of the app being used, launched as part of the latest version of Voice Access. IconNet can detect 31 different icon types (to be extended to more than 70 types soon) based on UI screenshots alone. IconNet is optimized to run on-device for mobile environments, with a compact size and fast inference time to enable a seamless user experience. The current IconNet model achieves a mean average precision (mAP) of 94.2% running at 9 FPS on a Pixel 3A.

Voice Access 5.0: the icons detected by IconNet can now be referred to by their names.

Detecting Icons in Screenshots
From a technical perspective, the problem of detecting icons on app screens is similar to classical object detection, in that individual elements are labelled by the model with their locations and sizes. But, in other ways, it’s quite different. Icons are typically small objects, with relatively basic geometric shapes and a limited range of colors, and app screens widely differ from natural images in that they are more structured and geometrical.

A significant challenge in the development of an on-device UI element detector for Voice Access is that it must be able to run on a wide variety of phones with a range of performance performance capabilities, while preserving the user’s privacy. For a fast user experience, a lightweight model with low inference latency is needed. Because Voice Access needs to use the labels in response to an utterance from a user (e.g., “tap camera”, or “show labels”) inference time needs to be short (<150 ms on a Pixel 3A) with a model size less than 10 MB.

IconNet
IconNet is based on the novel CenterNet architecture, which extracts features from input images and then predicts appropriate bounding box centers and sizes (in the form of heatmaps). CenterNet is particularly suited here because UI elements consist of simple, symmetric geometric shapes, making it easier to identify their centers than for natural images. The total loss used is a combination of a standard L1 loss for the icon sizes and a modified CornerNet Focal loss for the center predictions, the latter of which addresses icon class imbalances between commonly occurring icons (e.g., arrow backward, menu, more, and star) and underrepresented icons (end call, delete, launch apps, etc.)..

After experimenting with several backbones (MobileNet, ResNet, UNet, etc), we selected the most promising server-side architecture — Hourglass — as a starting point for designing a backbone tailored for icon and UI element detection. While this architecture is perfectly suitable for server side models, vanilla Hourglass backbones are not an option for a model that will run on a mobile device, due to their large size and slow inference time. We restricted our on-device network design to a single stack, and drastically reduced the width of the backbone. Furthermore, as the detection of icons relies on more local features (compared to real objects), we could further reduce the depth of the backbone without adversely affecting the performance. Ablation studies convinced us of the importance of skip connections and high resolution features. For example, trimming skip connections in the final layer reduced the mAP by 1.5%, and removing such connections from both the final and penultimate layers resulted in a decline of 3.5% mAP.

IconNet analyzes the pixels of the screen and identifies the centers of icons by generating heatmaps, which provide precise information about the position and type of the different types of icons present on the screen. This enables Voice Access users to refer to these elements by their name (e.g., “Tap ‘menu”).

Model Improvements
Once the backbone architecture was selected, we used neural architecture search (NAS) to explore variations on the network architecture and uncover an optimal set of training and model parameters that would balance model performance (mAP) with latency (FLOPs). Additionally, we used Fine-Grained Stochastic Architecture Search (FiGS) to further refine the backbone design. FiGS is a differentiable architecture search technique that uncovers sparse structures by pruning a candidate architecture and discarding unnecessary connections. This technique allowed us to reduce the model size by 20% without any loss in performance, and by 50% with only a minor drop of 0.3% in mAP.

Improving the quality of the training dataset also played an important role in boosting the model performance. We collected and labeled more than 700K screenshots, and in the process, we streamlined data collection by using heuristics and auxiliary models to identify rarer icons. We also took advantage of data augmentation techniques by enriching existing screenshots with infrequent icons.

To improve the inference time, we modified our model to run using Neural Networks API (NNAPI) on a variety of Qualcomm DSPs available on many mobile phones. For this we converted the model to use 8-bit integer quantization which gives the additional benefit of model size reduction. After some experimentation, we used quantization aware training to quantize the model, while matching the performance of a server-side floating point model. The quantized model results in a 6x speed-up (700ms vs 110ms) and 50% size reduction while losing only ~0.5% mAP compared to the unquantized model.

Results
We use traditional object detection metrics (e.g., mAP) to measure model performance. In addition, to better capture the use case of voice controlled user actions, we define a modified version of a false positive (FP) detection, where we penalize more incorrect detections for icon classes that are present on the screen. For comparing detections with ground truth, we use the center in region of interest (CIROI), another metric we developed for this work, which returns in a positive match when the center of the detected bounding box lies inside the ground truth bounding box. This better captures the Voice Access mode of operation, where actions are performed by tapping anywhere in the region of the UI element of interest.

We compared the IconNet model with various other mobile compatible object detectors, including MobileNetEdgeTPU and SSD MobileNet v2. Experiments showed that for a fixed latency, IconNet outperformed the other models in terms of [email protected] on our internal evaluation set.

Model    [email protected]
IconNet (Hourglass)    96%
IconNet (HRNet)    89%
MobilenetEdgeTPU (AutoML)    91%
SSD Mobilenet v2    88%

The performance advantage of IconNet persists when considering quantized models and models for a fixed latency budget.

Models (Quantized)    [email protected]    Model size    Latency*
IconNet (Currently deployed)    94.20%    8.5 MB    107 ms
IconNet (XS)    92.80%    2.3 MB    102 ms
IconNet (S)    91.70%    4.4 MB    45 ms
MobilenetEdgeTPU (AutoML)    88.90%    7.8 MB    26 ms
*Measured on Pixel 3A.

Conclusion and Future Work
We are constantly working on improving IconNet. Among other things, we are interested in increasing the range of elements supported by IconNet to include any generic UI element, such as images, text, or buttons. We also plan to extend IconNet to differentiate between similar looking icons by identifying their functionality. On the application side, we are hoping to increase the number of apps with valid content descriptions by augmenting developer tools to suggest content descriptions for different UI elements when building applications.

Acknowledgements
This project is the result of joint work with Maria Wang, Tautvydas Misiūnas, Lijuan Liu, Ying Xu, Nevan Wichers, Xiaoxue Zang, Gabriel Schubiner, Abhinav Rastogi, Jindong (JD) Chen, Abhanshu Sharma, Pranav Khaitan, Matt Sharifi and Blaise Aguera y Arcas. We sincerely thank our collaborators Robert Berry, Folawiyo Campbell, Shraman Ray Chaudhuri, Nghi Doan, Elad Eban, Marybeth Fair, Alec Go, Sahil Goel, Tom Hume, Cassandra Luongo, Yair Movshovitz-Attias, James Stout, Gabriel Taubman and Anton Vayvod. We are very grateful to Tom Small for assisting us in preparing the post.

Source: Google AI Blog


Improving Mobile App Accessibility with Icon Detection

Voice Access enables users to control their Android device hands free, using only verbal commands. In order to function properly, it needs on-screen user interface (UI) elements to have reliable accessibility labels, which are provided to the operating system’s accessibility services via the accessibility tree. Unfortunately, in many apps, adequate labels aren’t always available for UI elements, e.g. images and icons, reducing the usability of Voice Access.

The Voice Access app extracts elements from the view hierarchy to localize and annotate various UI elements. It can provide a precise description for elements that have an explicit content description. On the other hand, the absence of content description can result in many unrecognized elements undermining the ability of Voice Access to function with some apps.

Addressing this challenge requires a system that can automatically detect icons using only the pixel values displayed on the screen, regardless of whether icons have been given suitable accessibility labels. What little research exists on this topic typically uses classifiers, sometimes combined with language models to infer classes and attributes from UI elements. However, these classifiers still rely on the accessibility tree to obtain bounding boxes for UI elements, and fail when appropriate labels do not exist.

Here, we describe IconNet, a vision-based object detection model that can automatically detect icons on the screen in a manner that is agnostic to the underlying structure of the app being used, launched as part of the latest version of Voice Access. IconNet can detect 31 different icon types (to be extended to more than 70 types soon) based on UI screenshots alone. IconNet is optimized to run on-device for mobile environments, with a compact size and fast inference time to enable a seamless user experience. The current IconNet model achieves a mean average precision (mAP) of 94.2% running at 9 FPS on a Pixel 3A.

Voice Access 5.0: the icons detected by IconNet can now be referred to by their names.

Detecting Icons in Screenshots
From a technical perspective, the problem of detecting icons on app screens is similar to classical object detection, in that individual elements are labelled by the model with their locations and sizes. But, in other ways, it’s quite different. Icons are typically small objects, with relatively basic geometric shapes and a limited range of colors, and app screens widely differ from natural images in that they are more structured and geometrical.

A significant challenge in the development of an on-device UI element detector for Voice Access is that it must be able to run on a wide variety of phones with a range of performance performance capabilities, while preserving the user’s privacy. For a fast user experience, a lightweight model with low inference latency is needed. Because Voice Access needs to use the labels in response to an utterance from a user (e.g., “tap camera”, or “show labels”) inference time needs to be short (<150 ms on a Pixel 3A) with a model size less than 10 MB.

IconNet
IconNet is based on the novel CenterNet architecture, which extracts features from input images and then predicts appropriate bounding box centers and sizes (in the form of heatmaps). CenterNet is particularly suited here because UI elements consist of simple, symmetric geometric shapes, making it easier to identify their centers than for natural images. The total loss used is a combination of a standard L1 loss for the icon sizes and a modified CornerNet Focal loss for the center predictions, the latter of which addresses icon class imbalances between commonly occurring icons (e.g., arrow backward, menu, more, and star) and underrepresented icons (end call, delete, launch apps, etc.)..

After experimenting with several backbones (MobileNet, ResNet, UNet, etc), we selected the most promising server-side architecture — Hourglass — as a starting point for designing a backbone tailored for icon and UI element detection. While this architecture is perfectly suitable for server side models, vanilla Hourglass backbones are not an option for a model that will run on a mobile device, due to their large size and slow inference time. We restricted our on-device network design to a single stack, and drastically reduced the width of the backbone. Furthermore, as the detection of icons relies on more local features (compared to real objects), we could further reduce the depth of the backbone without adversely affecting the performance. Ablation studies convinced us of the importance of skip connections and high resolution features. For example, trimming skip connections in the final layer reduced the mAP by 1.5%, and removing such connections from both the final and penultimate layers resulted in a decline of 3.5% mAP.

IconNet analyzes the pixels of the screen and identifies the centers of icons by generating heatmaps, which provide precise information about the position and type of the different types of icons present on the screen. This enables Voice Access users to refer to these elements by their name (e.g., “Tap ‘menu”).

Model Improvements
Once the backbone architecture was selected, we used neural architecture search (NAS) to explore variations on the network architecture and uncover an optimal set of training and model parameters that would balance model performance (mAP) with latency (FLOPs). Additionally, we used Fine-Grained Stochastic Architecture Search (FiGS) to further refine the backbone design. FiGS is a differentiable architecture search technique that uncovers sparse structures by pruning a candidate architecture and discarding unnecessary connections. This technique allowed us to reduce the model size by 20% without any loss in performance, and by 50% with only a minor drop of 0.3% in mAP.

Improving the quality of the training dataset also played an important role in boosting the model performance. We collected and labeled more than 700K screenshots, and in the process, we streamlined data collection by using heuristics and auxiliary models to identify rarer icons. We also took advantage of data augmentation techniques by enriching existing screenshots with infrequent icons.

To improve the inference time, we modified our model to run using Neural Networks API (NNAPI) on a variety of Qualcomm DSPs available on many mobile phones. For this we converted the model to use 8-bit integer quantization which gives the additional benefit of model size reduction. After some experimentation, we used quantization aware training to quantize the model, while matching the performance of a server-side floating point model. The quantized model results in a 6x speed-up (700ms vs 110ms) and 50% size reduction while losing only ~0.5% mAP compared to the unquantized model.

Results
We use traditional object detection metrics (e.g., mAP) to measure model performance. In addition, to better capture the use case of voice controlled user actions, we define a modified version of a false positive (FP) detection, where we penalize more incorrect detections for icon classes that are present on the screen. For comparing detections with ground truth, we use the center in region of interest (CIROI), another metric we developed for this work, which returns in a positive match when the center of the detected bounding box lies inside the ground truth bounding box. This better captures the Voice Access mode of operation, where actions are performed by tapping anywhere in the region of the UI element of interest.

We compared the IconNet model with various other mobile compatible object detectors, including MobileNetEdgeTPU and SSD MobileNet v2. Experiments showed that for a fixed latency, IconNet outperformed the other models in terms of [email protected] on our internal evaluation set.

Model    [email protected]
IconNet (Hourglass)    96%
IconNet (HRNet)    89%
MobilenetEdgeTPU (AutoML)    91%
SSD Mobilenet v2    88%

The performance advantage of IconNet persists when considering quantized models and models for a fixed latency budget.

Models (Quantized)    [email protected]    Model size    Latency*
IconNet (Currently deployed)    94.20%    8.5 MB    107 ms
IconNet (XS)    92.80%    2.3 MB    102 ms
IconNet (S)    91.70%    4.4 MB    45 ms
MobilenetEdgeTPU (AutoML)    88.90%    7.8 MB    26 ms
*Measured on Pixel 3A.

Conclusion and Future Work
We are constantly working on improving IconNet. Among other things, we are interested in increasing the range of elements supported by IconNet to include any generic UI element, such as images, text, or buttons. We also plan to extend IconNet to differentiate between similar looking icons by identifying their functionality. On the application side, we are hoping to increase the number of apps with valid content descriptions by augmenting developer tools to suggest content descriptions for different UI elements when building applications.

Acknowledgements
This project is the result of joint work with Maria Wang, Tautvydas Misiūnas, Lijuan Liu, Ying Xu, Nevan Wichers, Xiaoxue Zang, Gabriel Schubiner, Abhinav Rastogi, Jindong (JD) Chen, Abhanshu Sharma, Pranav Khaitan, Matt Sharifi and Blaise Aguera y Arcas. We sincerely thank our collaborators Robert Berry, Folawiyo Campbell, Shraman Ray Chaudhuri, Nghi Doan, Elad Eban, Marybeth Fair, Alec Go, Sahil Goel, Tom Hume, Cassandra Luongo, Yair Movshovitz-Attias, James Stout, Gabriel Taubman and Anton Vayvod. We are very grateful to Tom Small for assisting us in preparing the post.

Source: Google AI Blog


Recognizing Pose Similarity in Images and Videos

Everyday actions, such as jogging, reading a book, pouring water, or playing sports, can be viewed as a sequence of poses, consisting of the position and orientation of a person’s body. An understanding of poses from images and videos is a crucial step for enabling a range of applications, including augmented reality display, full-body gesture control, and physical exercise quantification. However, a 3-dimensional pose captured in two dimensions in images and videos appears different depending on the viewpoint of the camera. The ability to recognize similarity in 3D pose using only 2D information will help vision systems better understand the world.

In “View-Invariant Probabilistic Embedding for Human Pose” (Pr-VIPE), a spotlight paper at ECCV 2020, we present a new algorithm for human pose perception that recognizes similarity in human body poses across different camera views by mapping 2D body pose keypoints to a view-invariant embedding space. This ability enables tasks, such as pose retrieval, action recognition, action video synchronization, and more. Compared to existing models that directly map 2D pose keypoints to 3D pose keypoints, the Pr-VIPE embedding space is (1) view-invariant, (2) probabilistic in order to capture 2D input ambiguity, and (3) does not require camera parameters during training or inference. Trained with in-lab setting data, the model works on in-the-wild images out of the box, given a reasonably good 2D pose estimator (e.g., PersonLab, BlazePose, among others). The model is simple, results in compact embeddings, and can be trained (in ~1 day) using 15 CPUs. We have released the code on our GitHub repo.

Pr-VIPE can be directly applied to align videos from different views.

Pr-VIPE
The input to Pr-VIPE is a set of 2D keypoints, from any 2D pose estimator that produces a minimum of 13 body keypoints, and the output is the mean and variance of the pose embedding. The distances between embeddings of 2D poses correlate to their similarities in absolute 3D pose space. Our approach is based on two observations:

  • The same 3D pose may appear very different in 2D as the viewpoint changes.
  • The same 2D pose can be projected from different 3D poses.

The first observation motivates the need for view-invariance. To accomplish this, we define the matching probability, i.e., the likelihood that different 2D poses were projected from the same, or similar 3D poses. The matching probability predicted by Pr-VIPE for matching pose pairs should be higher than for non-matching pairs.

To address the second observation, Pr-VIPE utilizes a probabilistic embedding formulation. Because many 3D poses can project to the same or similar 2D poses, the model input exhibits an inherent ambiguity that is difficult to capture through deterministic mapping point-to-point in embedding space. Therefore, we map a 2D pose through a probabilistic mapping to an embedding distribution, of which we use the variance to represent the uncertainty of the input 2D pose. As an example, in the figure below the third 2D view of the 3D pose on the left is similar to the first 2D view of a different 3D pose on the right, so we map them into a similar location in the embedding space with large variances.

Pr-VIPE enables vision systems to recognize 2D poses across views. We embed 2D poses using Pr-VIPE such that the embeddings are (1) view-invariant (2D projections of similar 3D poses are embedded close together) and (2) probabilistic. By embedding detected 2D poses, Pr-VIPE enables direct retrieval of pose images from different views, and can also be applied to action recognition and video alignment.
View-Invariance
During training, we use 2D poses from two sources: multi-view images and projections of groundtruth 3D poses. Triplets of 2D poses (anchor, positive, and negative) are selected from a batch, where the anchor and positive are two different projections of the same 3D pose, and the negative is a projection of a non-matching 3D pose. Pr-VIPE then estimates the matching probability of 2D pose pairs from their embeddings.
During training, we push the matching probability of positive pairs to be close to 1 with a positive pairwise loss in which we minimize the embedding distance between positive pairs, and the matching probability of negative pairs to be small by maximizing the ratio of the matching probabilities between positive and negative pairs with a triplet ratio loss.
Overview of the Pr-VIPE model. During training, we apply three losses (triplet ratio loss, positive pairwise loss, and a prior loss that applies a unit Gaussian prior to our embeddings). During inference, the model maps an input 2D pose to a probabilistic, view-invariant embedding.
Probabilistic Embedding
Pr-VIPE maps a 2D pose to a probabilistic embedding as a multivariate Gaussian distribution using a sampling-based approach for similarity score computation between two distributions. During training, we use a Gaussian prior loss to regularize the predicted distribution.

Evaluation
We propose a new cross-view pose retrieval benchmark to evaluate the view-invariance property of the embedding. Given a monocular pose image, cross-view retrieval aims to retrieve the same pose from different views without using camera parameters. The results demonstrate that Pr-VIPE retrieves poses more accurately across views compared to baseline methods in both evaluated datasets (Human3.6M, MPI-INF-3DHP).

Pr-VIPE retrieves poses across different views more accurately relative to the baseline method (3D pose estimation).

Common 3D pose estimation methods (such as the simple baseline used for comparison above, SemGCN, and EpipolarPose, amongst many others), predict 3D poses in camera coordinates, which are not directly view-invariant. Thus, rigid alignment between every query-index pair is required for retrieval using estimated 3D poses, which is computationally expensive due to the need for singular value decomposition (SVD). In contrast, Pr-VIPE embeddings can be directly used for distance computation in Euclidean space, without any post-processing.

Applications
View-invariant pose embedding can be applied to many image and video related tasks. Below, we show Pr-VIPE applied to cross-view retrieval on in-the-wild images without using camera parameters.


We can retrieve in-the-wild images from different views without using camera parameters by embedding the detected 2D pose using Pr-VIPE. Using the query image (top row), we search for a matching pose from a different camera view and we show the nearest neighbor retrieval (bottom row). This enables us to search for matching poses across camera views more easily.

The same Pr-VIPE model can also be used for video alignment. To do so, we stack Pr-VIPE embeddings within a small time window, and use the dynamic time warping (DTW) algorithm to align video pairs.

Manual video alignment is difficult and time-consuming. Here, Pr-VIPE is applied to automatically align videos of the same action repeated from different views.

The video alignment distance calculated via DTW can then be used for action recognition by classifying videos using nearest neighbor search. We evaluate the Pr-VIPE embedding using the Penn Action dataset and demonstrate that using the Pr-VIPE embedding without fine-tuning on the target dataset, yields highly competitive recognition accuracy. In addition, we show that Pr-VIPE even achieves relatively accurate results using only videos from a single view in the index set.

Pr-VIPE recognizes action across views using pose inputs only, and is comparable to or better than methods using pose only or with additional context information (such as Iqbal et al., Liu and Yuan, Luvizon et al., and Du et al.). When action labels are only available for videos from a single view, Pr-VIPE (1-view only) can still achieve relatively accurate results.

Conclusion
We introduce the Pr-VIPE model for mapping 2D human poses to a view-invariant probabilistic embedding space, and show that the learned embeddings can be directly used for pose retrieval, action recognition, and video alignment. Our cross-view retrieval benchmark can be used to test the view-invariant property of other embeddings. We look forward to hearing about what you can do with pose embeddings!

Acknowledgments
Special thanks to Jiaping Zhao, Liang-Chieh Chen, Long Zhao (Rutgers University), Liangzhe Yuan, Yuxiao Wang, Florian Schroff, Hartwig Adam, and the Mobile Vision team for the wonderful collaboration and support.

Source: Google AI Blog


MediaPipe Holistic — Simultaneous Face, Hand and Pose Prediction, on Device

Real-time, simultaneous perception of human pose, face landmarks and hand tracking on mobile devices can enable a variety of impactful applications, such as fitness and sport analysis, gesture control and sign language recognition, augmented reality effects and more. MediaPipe, an open-source framework designed specifically for complex perception pipelines leveraging accelerated inference (e.g., GPU or CPU), already offers fast and accurate, yet separate, solutions for these tasks. Combining them all in real-time into a semantically consistent end-to-end solution is a uniquely difficult problem requiring simultaneous inference of multiple, dependent neural networks.

Today, we are excited to announce MediaPipe Holistic, a solution to this challenge that provides a novel state-of-the-art human pose topology that unlocks novel use cases. MediaPipe Holistic consists of a new pipeline with optimized pose, face and hand components that each run in real-time, with minimum memory transfer between their inference backends, and added support for interchangeability of the three components, depending on the quality/speed tradeoffs. When including all three components, MediaPipe Holistic provides a unified topology for a groundbreaking 540+ keypoints (33 pose, 21 per-hand and 468 facial landmarks) and achieves near real-time performance on mobile devices. MediaPipe Holistic is being released as part of MediaPipe and is available on-device for mobile (Android, iOS) and desktop. We are also introducing MediaPipe’s new ready-to-use APIs for research (Python) and web (JavaScript) to ease access to the technology.

Top: MediaPipe Holistic results on sport and dance use-cases. Bottom: “Silence” and “Hello” gestures. Note, that our solution consistently identifies a hand as either right (blue color) or left (orange color).

Pipeline and Quality
The MediaPipe Holistic pipeline integrates separate models for pose, face and hand components, each of which are optimized for their particular domain. However, because of their different specializations, the input to one component is not well-suited for the others. The pose estimation model, for example, takes a lower, fixed resolution video frame (256x256) as input. But if one were to crop the hand and face regions from that image to pass to their respective models, the image resolution would be too low for accurate articulation. Therefore, we designed MediaPipe Holistic as a multi-stage pipeline, which treats the different regions using a region appropriate image resolution.

First, MediaPipe Holistic estimates the human pose with BlazePose’s pose detector and subsequent keypoint model. Then, using the inferred pose key points, it derives three regions of interest (ROI) crops for each hand (2x) and the face, and employs a re-crop model to improve the ROI (details below). The pipeline then crops the full-resolution input frame to these ROIs and applies task-specific face and hand models to estimate their corresponding keypoints. Finally, all key points are merged with those of the pose model to yield the full 540+ keypoints.

MediaPipe Holistic pipeline overview.

To streamline the identification of ROIs, a tracking approach similar to the one used for the standalone face and hand pipelines is utilized. This approach assumes that the object doesn't move significantly between frames, using an estimation from the previous frame as a guide to the object region in the current one. However, during fast movements, the tracker can lose the target, which requires the detector to re-localize it in the image. MediaPipe Holistic uses pose prediction (on every frame) as an additional ROI prior to reduce the response time of the pipeline when reacting to fast movements. This also enables the model to retain semantic consistency across the body and its parts by preventing a mixup between left and right hands or body parts of one person in the frame with another.

In addition, the resolution of the input frame to the pose model is low enough that the resulting ROIs for face and hands are still too inaccurate to guide the re-cropping of those regions, which require a precise input crop to remain lightweight. To close this accuracy gap we use lightweight face and hand re-crop models that play the role of spatial transformers and cost only ~10% of the corresponding model's inference time.

 MEH   FLE 
 Tracking pipeline (baseline)   9.8%   3.1% 
 Pipeline without re-crops   11.8%   3.5% 
 Pipeline with re-crops   9.7%   3.1% 
Hand prediction quality.The mean error per hand (MEH) is normalized by the hand size. The face landmarks error (FLE) is normalized by the inter-pupillary distance.

Performance
MediaPipe Holistic requires coordination between up to 8 models per frame — 1 pose detector, 1 pose landmark model, 3 re-crop models and 3 keypoint models for hands and face. While building this solution, we optimized not only machine learning models, but also pre- and post-processing algorithms (e.g., affine transformations), which take significant time on most devices due to pipeline complexity. In this case, moving all the pre-processing computations to GPU resulted in ~1.5 times overall pipeline speedup depending on the device. As a result, MediaPipe Holistic runs in near real-time performance even on mid-tier devices and in the browser.

 Phone   FPS 
 Google Pixel 2 XL   18 
 Samsung S9+   20 
 15-inch MacBook Pro 2017   15 
Performance on various mid-tier devices, measured in frames per second (FPS) using TFLite GPU.

The multi-stage nature of the pipeline provides two more performance benefits. As models are mostly independent, they can be replaced with lighter or heavier versions (or turned off completely) depending on the performance and accuracy requirements. Also, once pose is inferred, one knows precisely whether hands and face are within the frame bounds, allowing the pipeline to skip inference on those body parts.

Applications
MediaPipe Holistic, with its 540+ key points, aims to enable a holistic, simultaneous perception of body language, gesture and facial expressions. Its blended approach enables remote gesture interfaces, as well as full-body AR, sports analytics, and sign language recognition. To demonstrate the quality and performance of the MediaPipe Holistic, we built a simple remote control interface that runs locally in the browser and enables a compelling user interaction, no mouse or keyboard required. The user can manipulate objects on the screen, type on a virtual keyboard while sitting on the sofa, and point to or touch specific face regions (e.g., mute or turn off the camera). Underneath it relies on accurate hand detection with subsequent gesture recognition mapped to a "trackpad" space anchored to the user’s shoulder, enabling remote control from up to 4 meters.

This technique for gesture control can unlock various novel use-cases when other human-computer interaction modalities are not convenient. Try it out in our web demo and prototype your own ideas with it.

In-browser touchless control demos. Left: Palm picker, touch interface, keyboard. Right: Distant touchless keyboard. Try it out!

MediaPipe for Research and Web
To accelerate ML research as well as its adoption in the web developer community, MediaPipe now offers ready-to-use, yet customizable ML solutions in Python and in JavaScript. We are starting with those in our previous publications: Face Mesh, Hands and Pose, including MediaPipe Holistic, with many more to come. Try them directly in the web browser: for Python using the notebooks in MediaPipe on Google Colab, and for JavaScript with your own webcam input in MediaPipe on CodePen!

Conclusion
We hope the release of MediaPipe Holistic will inspire the research and development community members to build new unique applications. We anticipate that these pipelines will open up avenues for future research into challenging domains, such as sign-language recognition, touchless control interfaces, or other complex use cases. We are looking forward to seeing what you can build with it!

Complex and dynamic hand gestures. Videos by Dr. Bill Vicars, used with permission.

Acknowledgments
Special thanks to all our team members who worked on the tech with us: Fan Zhang, Gregory Karpiak, Kanstantsin Sokal, Juhyun Lee, Hadon Nash, Chuo-Ling Chang, Jiuqiang Tang, Nikolay Chirkov, Camillo Lugaresi, George Sung, Michael Hays, Tyler Mullen, Chris McClanahan, Ekaterina Ignasheva, Marat Dukhan, Artsiom Ablavatski, Yury Kartynnik, Karthik Raveendran, Andrei Vakunov, Andrei Tkachenka, Suril Shah, Buck Bourdon, Ming Guang Yong, Esha Uboweja, Siarhei Kazakou, Andrei Kulik, Matsvei Zhdanovich, and Matthias Grundmann.

Source: Google AI Blog


Transformers for Image Recognition at Scale

While convolutional neural networks (CNNs) have been used in computer vision since the 1980s, they were not at the forefront until 2012 when AlexNet surpassed the performance of contemporary state-of-the-art image recognition methods by a large margin. Two factors helped enable this breakthrough: (i) the availability of training sets like ImageNet, and (ii) the use of commoditized GPU hardware, which provided significantly more compute for training. As such, since 2012, CNNs have become the go-to model for vision tasks.

The benefit of using CNNs was that they avoided the need for hand-designed visual features, instead learning to perform tasks directly from data “end to end”. However, while CNNs avoid hand-crafted feature-extraction, the architecture itself is designed specifically for images and can be computationally demanding. Looking forward to the next generation of scalable vision models, one might ask whether this domain-specific design is necessary, or if one could successfully leverage more domain agnostic and computationally efficient architectures to achieve state-of-the-art results.

As a first step in this direction, we present the Vision Transformer (ViT), a vision model based as closely as possible on the Transformer architecture originally designed for text-based tasks. ViT represents an input image as a sequence of image patches, similar to the sequence of word embeddings used when applying Transformers to text, and directly predicts class labels for the image. ViT demonstrates excellent performance when trained on sufficient data, outperforming a comparable state-of-the-art CNN with four times fewer computational resources. To foster additional research in this area, we have open-sourced both the code and models.

The Vision Transformer treats an input image as a sequence of patches, akin to a series of word embeddings generated by a natural language processing (NLP) Transformer.

The Vision Transformer
The original text Transformer takes as input a sequence of words, which it then uses for classification, translation, or other NLP tasks. For ViT, we make the fewest possible modifications to the Transformer design to make it operate directly on images instead of words, and observe how much about image structure the model can learn on its own.

ViT divides an image into a grid of square patches. Each patch is flattened into a single vector by concatenating the channels of all pixels in a patch and then linearly projecting it to the desired input dimension. Because Transformers are agnostic to the structure of the input elements we add learnable position embeddings to each patch, which allow the model to learn about the structure of the images. A priori, ViT does not know about the relative location of patches in the image, or even that the image has a 2D structure — it must learn such relevant information from the training data and encode structural information in the position embeddings.

Scaling Up

We first train ViT on ImageNet, where it achieves a best score of 77.9% top-1 accuracy. While this is decent for a first attempt, it falls far short of the state of the art — the current best CNN trained on ImageNet with no extra data reaches 85.8%. Despite mitigation strategies (e.g., regularization), ViT overfits the ImageNet task due to its lack of inbuilt knowledge about images.

To investigate the impact of dataset size on model performance, we train ViT on ImageNet-21k (14M images, 21k classes) and JFT (300M images, 18k classes), and compare the results to a state-of-the-art CNN, Big Transfer (BiT), trained on the same datasets. As previously observed, ViT performs significantly worse than the CNN equivalent (BiT) when trained on ImageNet (1M images). However, on ImageNet-21k (14M images) performance is comparable, and on JFT (300M images), ViT now outperforms BiT.

Finally, we investigate the impact of the amount of computation involved in training the models. For this, we train several different ViT models and CNNs on JFT. These models span a range of model sizes and training durations. As a result, they require varying amounts of compute for training. We observe that, for a given amount of compute, ViT yields better performance than the equivalent CNNs.

Left: Performance of ViT when pre-trained on different datasets. Right: ViT yields a good performance/compute trade-off.

High-Performing Large-Scale Image Recognition
Our data suggest that (1) with sufficient training ViT can perform very well, and (2) ViT yields an excellent performance/compute trade-off at both smaller and larger compute scales. Therefore, to see if performance improvements carried over to even larger scales, we trained a 600M-parameter ViT model.

This large ViT model attains state-of-the-art performance on multiple popular benchmarks, including 88.55% top-1 accuracy on ImageNet and 99.50% on CIFAR-10. ViT also performs well on the cleaned-up version of the ImageNet evaluations set “ImageNet-Real”, attaining 90.72% top-1 accuracy. Finally, ViT works well on diverse tasks, even with few training data points. For example, on the VTAB-1k suite (19 tasks with 1,000 data points each), ViT attains 77.63%, significantly ahead of the single-model state of the art (SOTA) (76.3%), and even matching SOTA attained by an ensemble of multiple models (77.6%). Most importantly, these results are obtained using fewer compute resources compared to previous SOTA CNNs, e.g., 4x fewer than the pre-trained BiT models.

Vision Transformer matches or outperforms state-of-the-art CNNs on popular benchmarks. Left: Popular image classification tasks (ImageNet, including new validation labels ReaL, and CIFAR, Pets, and Flowers). Right: Average across 19 tasks in the VTAB classification suite.

Visualizations
To gain some intuition into what the model learns, we visualize some of its internal workings. First, we look at the position embeddings — parameters that the model learns to encode the relative location of patches — and find that ViT is able to reproduce an intuitive image structure. Each position embedding is most similar to others in the same row and column, indicating that the model has recovered the grid structure of the original images. Second, we examine the average spatial distance between one element attending to another for each transformer block. At higher layers (depths of 10-20) only global features are used (i.e., large attention distances), but the lower layers (depths 0-5) capture both global and local features, as indicated by a large range in the mean attention distance. By contrast, only local features are present in the lower layers of a CNN. These experiments indicate that ViT can learn features hard-coded into CNNs (such as awareness of grid structure), but is also free to learn more generic patterns, such as a mix of local and global features at lower layers, that can aid generalization.

Left: ViT learns the grid like structure of the image patches via its position embeddings. Right: The lower layers of ViT contain both global and local features, the higher layers contain only global features.

Summary
While CNNs have revolutionized computer vision, our results indicate that models tailor-made for imaging tasks may be unnecessary, or even sub-optimal. With ever-increasing dataset sizes, and the continued development of unsupervised and semi-supervised methods, the development of new vision architectures that train more efficiently on these datasets becomes increasingly important. We believe ViT is a preliminary step towards generic, scalable architectures that can solve many vision tasks, or even tasks from many domains, and are excited for future developments.

A preprint of our work as well as code and models are publically available.

Acknowledgements
We would like to thank our co-authors in Berlin, Zürich, and Amsterdam: Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, and Jakob Uszkoreit. We would like to thank Andreas Steiner for crucial help with infrastructure and open-sourcing, Joan Puigcerver and Maxim Neumann for work on large-scale training infrastructure, and Dmitry Lepikhin, Aravindh Mahendran, Daniel Keysers, Mario Lučić, Noam Shazeer, and Colin Raffel for useful discussions. Finally, we thank Tom Small for creating the Visual Transformer animation in this post.

Source: Google AI Blog


Using GANs to Create Fantastical Creatures

Creating art for digital video games takes a high degree of artistic creativity and technical knowledge, while also requiring game artists to quickly iterate on ideas and produce a high volume of assets, often in the face of tight deadlines. What if artists had a paintbrush that acted less like a tool and more like an assistant? A machine learning model acting as such a paintbrush could reduce the amount of time necessary to create high-quality art without sacrificing artistic choices, perhaps even enhancing creativity.

Today, we present Chimera Painter, a trained machine learning (ML) model that automatically creates a fully fleshed out rendering from a user-supplied creature outline. Employed as a demo application, Chimera Painter adds features and textures to a creature outline segmented with body part labels, such as “wings” or “claws”, when the user clicks the “transform” button. Below is an example using the demo with one of the preset creature outlines.

Using an image imported to Chimera Painter or generated with the tools provided, an artist can iteratively construct or modify a creature outline and use the ML model to generate realistic looking surface textures. In this example, an artist (Lee Dotson) customizes one of the creature designs that comes pre-loaded in the Chimera Painter demo.

In this post, we describe some of the challenges in creating the ML model behind Chimera Painter and demonstrate how one might use the tool for the creation of video game-ready assets.

Prototyping for a New Type of Model
In developing an ML model to produce video-game ready creature images, we created a digital card game prototype around the concept of combining creatures into new hybrids that can then battle each other. In this game, a player would begin with cards of real-world animals (e.g., an axolotl or a whale) and could make them more powerful by combining them (making the dreaded Axolotl-Whale chimera). This provided a creative environment for demonstrating an image-generating model, as the number of possible chimeras necessitated a method for quickly designing large volumes of artistic assets that could be combined naturally, while still retaining identifiable visual characteristics of the original creatures.

Since our goal was to create high-quality creature card images guided by artist input, we experimented with generative adversarial networks (GANs), informed by artist feedback, to create creature images that would be appropriate for our fantasy card game prototype. GANs pair two convolutional neural networks against each other: a generator network to create new images and a discriminator network to determine if these images are samples from the training dataset (in this case, artist-created images) or not. We used a variant called a conditional GAN, where the generator takes a separate input to guide the image generation process. Interestingly, our approach was a strict departure from other GAN efforts, which typically focus on photorealism.

To train the GANs, we created a dataset of full color images with single-species creature outlines adapted from 3D creature models. The creature outlines characterized the shape and size of each creature, and provided a segmentation map that identified individual body parts. After model training, the model was tasked with generating multi-species chimeras, based on outlines provided by artists. The best performing model was then incorporated into Chimera Painter. Below we show some sample assets generated using the model, including single-species creatures, as well as the more complex multi-species chimeras.

Generated card art integrated into the card game prototype showing basic creatures (bottom row) and chimeras from multiple creatures, including an Antlion-Porcupine, Axolotl-Whale, and a Crab-Antion-Moth (top row). More info about the game itself is detailed in this Stadia Research presentation.

Learning to Generate Creatures with Structure
An issue with using GANs for generating creatures was the potential for loss of anatomical and spatial coherence when rendering subtle or low-contrast parts of images, despite these being of high perceptual importance to humans. Examples of this can include eyes, fingers, or even distinguishing between overlapping body parts with similar textures (see the affectionately named BoggleDog below).

GAN-generated image showing mismatched body parts.

Generating chimeras required a new non-photographic fantasy-styled dataset with unique characteristics, such as dramatic perspective, composition, and lighting. Existing repositories of illustrations were not appropriate to use as datasets for training an ML model, because they may be subject to licensing restrictions, have conflicting styles, or simply lack the variety needed for this task.

To solve this, we developed a new artist-led, semi-automated approach for creating an ML training dataset from 3D creature models, which allowed us to work at scale and rapidly iterate as needed. In this process, artists would create or obtain a set of 3D creature models, one for each creature type needed (such as hyenas or lions). Artists then produced two sets of textures that were overlaid on the 3D model using the Unreal Engine — one with the full color texture (left image, below) and the other with flat colors for each body part (e.g., head, ears, neck, etc), called a “segmentation map” (right image, below). This second set of body part segments was given to the model at training to ensure that the GAN learned about body part-specific structure, shapes, textures, and proportions for a variety of creatures.

Example dataset training image and its paired segmentation map.

The 3D creature models were all placed in a simple 3D scene, again using the Unreal Engine. A set of automated scripts would then take this 3D scene and interpolate between different poses, viewpoints, and zoom levels for each of the 3D creature models, creating the full color images and segmentation maps that formed the training dataset for the GAN. Using this approach, we generated 10,000+ image + segmentation map pairs per 3D creature model, saving the artists millions of hours of time compared to creating such data manually (at approximately 20 minutes per image).

Fine Tuning
The GAN had many different hyper-parameters that could be adjusted, leading to different qualities in the output images. In order to better understand which versions of the model were better than others, artists were provided samples for different creature types generated by these models and asked to cull them down to a few best examples. We gathered feedback about desired characteristics present in these examples, such as a feeling of depth, style with regard to creature textures, and realism of faces and eyes. This information was used both to train new versions of the model and, after the model had generated hundreds of thousands of creature images, to select the best image from each creature category (e.g., gazelle, lynx, gorilla, etc).

We tuned the GAN for this task by focusing on the perceptual loss. This loss function component (also used in Stadia’s Style Transfer ML) computes a difference between two images using extracted features from a separate convolutional neural network (CNN) that was previously trained on millions of photographs from the ImageNet dataset. The features are extracted from different layers of the CNN and a weight is applied to each, which affects their contribution to the final loss value. We discovered that these weights were critically important in determining what a final generated image would look like. Below are some examples from the GAN trained with different perceptual loss weights.

Dino-Bat Chimeras generated using varying perceptual loss weights.

Some of the variation in the images above is due to the fact that the dataset includes multiple textures for each creature (for example, a reddish or grayish version of the bat). However, ignoring the coloration, many differences are directly tied to changes in perceptual loss values. In particular, we found that certain values brought out sharper facial features (e.g., bottom right vs. top right) or “smooth” versus “patterned” (top right vs. bottom left) that made generated creatures feel more real.

Here are some creatures generated from the GAN trained with different perceptual loss weights, showing off a small sample of the outputs and poses that the model can handle.

Creatures generated using different models.
A generated chimera (Dino-Bat-Hyena, to be exact) created using the conditional GAN. Output from the GAN (left) and the post-processed / composited card (right).

Chimera Painter
The trained GAN is now available in the Chimera Painter demo, allowing artists to work iteratively with the model, rather than drawing dozens of similar creatures from scratch. An artist can select a starting point and then adjust the shape, type, or placement of creature parts, enabling rapid exploration and for the creation of a large volume of images. The demo also allows for uploading a creature outline created in an external program, like Photoshop. Simply download one of the preset creature outlines to get the colors needed for each creature part and use this as a template for drawing one outside of Chimera Painter, and then use the “Load’ button on the demo to use this outline to flesh out your creation.

It is our hope that these GAN models and the Chimera Painter demonstration tool might inspire others to think differently about their art pipeline. What can one create when using machine learning as a paintbrush?

Acknowledgments
This project is conducted in collaboration with many people. Thanks to Ryan Poplin, Lee Dotson, Trung Le, Monica Dinculescu, Marc Destefano, Aaron Cammarata, Maggie Oh, Richard Wu, Ji Hun Kim, Erin Hoffman-John, and Colin Boswell. Thanks to everyone who pitched in to give hours of art direction, technical feedback, and drawings of fantastic creatures.

Source: Google AI Blog


Announcing the Objectron Dataset

The state of the art in machine learning (ML) has achieved exceptional accuracy on many computer vision tasks solely by training models on photos. Building upon these successes and advancing 3D object understanding has great potential to power a wider range of applications, such as augmented reality, robotics, autonomy, and image retrieval. For example, earlier this year we released MediaPipe Objectron, a set of real-time 3D object detection models designed for mobile devices, which were trained on a fully annotated, real-world 3D dataset, that can predict objects’ 3D bounding boxes.

Yet, understanding objects in 3D remains a challenging task due to the lack of large real-world datasets compared to 2D tasks (e.g., ImageNet, COCO, and Open Images). To empower the research community for continued advancement in 3D object understanding, there is a strong need for the release of object-centric video datasets, which capture more of the 3D structure of an object, while matching the data format used for many vision tasks (i.e., video or camera streams), to aid in the training and benchmarking of machine learning models.

Today, we are excited to release the Objectron dataset, a collection of short, object-centric video clips capturing a larger set of common objects from different angles. Each video clip is accompanied by AR session metadata that includes camera poses and sparse point-clouds. The data also contain manually annotated 3D bounding boxes for each object, which describe the object’s position, orientation, and dimensions. The dataset consists of 15K annotated video clips supplemented with over 4M annotated images collected from a geo-diverse sample (covering 10 countries across five continents).

Example videos in the Objectron dataset.

A 3D Object Detection Solution
Along with the dataset, we are also sharing a 3D object detection solution for four categories of objects — shoes, chairs, mugs, and cameras. These models are released in MediaPipe, Google's open source framework for cross-platform customizable ML solutions for live and streaming media, which also powers ML solutions like on-device real-time hand, iris and body pose tracking.

Sample results of 3D object detection solution running on mobile.

In contrast to the previously released single-stage Objectron model, these newest versions utilize a two-stage architecture. The first stage employs the TensorFlow Object Detection model to find the 2D crop of the object. The second stage then uses the image crop to estimate the 3D bounding box while simultaneously computing the 2D crop of the object for the next frame, so that the object detector does not need to run every frame. The second stage 3D bounding box predictor runs at 83 FPS on Adreno 650 mobile GPU.

Diagram of a reference 3D object detection solution.

Evaluation Metric for 3D Object Detection
With ground truth annotations, we evaluate the performance of 3D object detection models using 3D intersection over union (IoU) similarity statistics, a commonly used metric for computer vision tasks, which measures how close the bounding boxes are to the ground truth.

We propose an algorithm for computing accurate 3D IoU values for general 3D-oriented boxes. First, we compute the intersection points between faces of the two boxes using Sutherland-Hodgman Polygon clipping algorithm. This is similar to frustum culling, a technique used in computer graphics. The volume of the intersection is computed by the convex hull of all the clipped polygons. Finally, the IoU is computed from the volume of the intersection and volume of the union of two boxes. We are releasing the evaluation metrics source code along with the dataset.

Compute the 3D intersection over union using the polygon clipping algorithm, Left: Compute the intersection points of each face by clipping the polygon against the box. Right: Compute the volume of intersection by computing the convex hull of all intersection points (green).

Dataset Format
The technical details of the Objectron dataset, including usage and tutorials, are available on the dataset website. The dataset includes bikes, books, bottles, cameras, cereal boxes, chairs, cups, laptops, and shoes, and is stored in the objectron bucket on Google Cloud storage with the following assets:

  • The video sequences
  • The annotation labels (3D bounding boxes for objects)
  • AR metadata (such as camera poses, point clouds, and planar surfaces)
  • Processed dataset: shuffled version of the annotated frames, in tf.example format for images and SequenceExample format for videos.
  • Supporting scripts to run evaluation based on the metric described above
  • Supporting scripts to load the data into Tensorflow, PyTorch, and Jax and to visualize the dataset, including “Hello World” examples

With the dataset, we are also open-sourcing a data-pipeline to parse the dataset in popular Tensorflow, PyTorch and Jax frameworks. Example colab notebooks are also provided.

By releasing this Objectron dataset, we hope to enable the research community to push the limits of 3D object geometry understanding. We also hope to foster new research and applications, such as view synthesis, improved 3D representation, and unsupervised learning. Stay tuned for future activities and developments by joining our mailing list and visiting our github page.

Acknowledgements
The research described in this post was done by Adel Ahmadyan, Liangkai Zhang, Jianing Wei, Artsiom Ablavatski, Mogan Shieh, Ryan Hickman, Buck Bourdon, Alexander Kanaukou, Chuo-Ling Chang, Matthias Grundmann, ‎and Tom Funkhouser. We thank Aliaksandr Shyrokau, Sviatlana Mialik, Anna Eliseeva, and the annotation team for their high quality annotations. We also would like to thank Jonathan Huang and Vivek Rathod for their guidance on TensorFlow Object Detection API.

Source: Google AI Blog