Search

Traditional computer vision models are trained to predict a fixed set of predefined categories. Recently, natural language has been shown to be a broader and richer source of supervision that provides finer descriptions to visual concepts than supervised "gold" labels. Previous works, such as CLIP, use a simple pretraining task of predicting the pairings between images and text captions. CLIP, however, is data hungry and requires more than 400M image text pairs for training. We propose a data-efficient contrastive distillation method that uses soft labels to learn from noisy image-text pairs. Our model transfers knowledge from pretrained image and sentence encoders and achieves strong performance with only 3M image text pairs, 133x smaller than CLIP. Our method exceeds the previous SoTA of general zero-shot learning on ImageNet 21k+1k by 73% relatively with a ResNet50 image encoder and DeCLUTR text encoder. We also beat CLIP by 10.5% relatively on zero-shot evaluation on Google Open Images (19,958 classes)., Comment: 4 pages, 1 figure

By the one-to-many nature of the super-resolution (SR) problem, a single low-resolution (LR) image can be mapped to many high-resolution (HR) images. However, learning based SR algorithms are trained to map an LR image to the corresponding ground truth (GT) HR image in the training dataset. The training loss will increase and penalize the algorithm when the output does not exactly match the GT target, even when the outputs are mathematically valid candidates according to the SR framework. This becomes more problematic for the blind SR, as diverse unknown blur kernels exacerbate the ill-posedness of the problem. To this end, we propose a fundamentally different approach for the SR by introducing the concept of the adaptive target. The adaptive target is generated from the original GT target by a transformation to match the output of the SR network. The adaptive target provides an effective way for the SR algorithm to deal with the ill-posed nature of the SR, by providing the algorithm with the flexibility of accepting a variety of valid solutions. Experimental results show the effectiveness of our algorithm, especially for improving the perceptual quality of HR outputs.

Self-attention is a corner stone for transformer models. However, our analysis shows that self-attention in vision transformer inference is extremely sparse. When applying a sparsity constraint, our experiments on image (ImageNet- 1K) and video (Kinetics-400) understanding show we can achieve 95% sparsity on the self-attention maps while main-taining the performance drop to be less than 2 points. This motivates us to rethink the role of self-attention in vision transformer models.

Neural Architecture Search (NAS) yields state-of-the-art neural networks that outperform their best manually-designed counterparts. However, previous NAS methods search for architectures under one set of training hyper-parameters (i.e., a training recipe), overlooking superior architecture-recipe combinations. To address this, we present Neural Architecture-Recipe Search (NARS) to search both (a) architectures and (b) their corresponding training recipes, simultaneously. NARS utilizes an accuracy predictor that scores architecture and training recipes jointly, guiding both sample selection and ranking. Furthermore, to compensate for the enlarged search space, we leverage "free" architecture statistics (e.g., FLOP count) to pretrain the predictor, significantly improving its sample efficiency and prediction reliability. After training the predictor via constrained iterative optimization, we run fast evolutionary searches in just CPU minutes to generate architecturerecipe pairs for a variety of resource constraints, called FBNetV3. FBNetV3 makes up a family of state-of-the-art compact neural networks that outperform both automatically and manually-designed competitors. For example, FB-NetV3 matches both EfficientNet and ResNeSt accuracy on ImageNet with up to 2.0× and 7.1 × fewer FLOPs, respectively. Furthermore, FBNetV3 yields significant performance gains for downstream object detection tasks, improving mAP despite 18% fewer FLOPs and 34% fewer parameters than EfficientNet-based equivalents.

Differentiable Neural Architecture Search (DNAS) has demonstrated great success in designing state-of-the-art, efficient neural networks. However, DARTS-based DNAS's search space is small when compared to other search methods', since all candidate network layers must be explicitly instantiated in memory. To address this bottleneck, we propose a memory and computationally efficient DNAS variant: DMaskingNAS. This algorithm expands the search space by up to $10^{14}\times$ over conventional DNAS, supporting searches over spatial and channel dimensions that are otherwise prohibitively expensive: input resolution and number of filters. We propose a masking mechanism for feature map reuse, so that memory and computational costs stay nearly constant as the search space expands. Furthermore, we employ effective shape propagation to maximize per-FLOP or per-parameter accuracy. The searched FBNetV2s yield state-of-the-art performance when compared with all previous architectures. With up to 421$\times$ less search cost, DMaskingNAS finds models with 0.9% higher accuracy, 15% fewer FLOPs than MobileNetV3-Small; and with similar accuracy but 20% fewer FLOPs than Efficient-B0. Furthermore, our FBNetV2 outperforms MobileNetV3 by 2.6% in accuracy, with equivalent model size. FBNetV2 models are open-sourced at https://github.com/facebookresearch/mobile-vision., Comment: 8 pages, 10 figures, accepted to CVPR 2020

3D photography is a new medium that allows viewers to more fully experience a captured moment. In this work, we refer to a 3D photo as one that displays parallax induced by moving the viewpoint (as opposed to a stereo pair with a fixed viewpoint). 3D photos are static in time, like traditional photos, but are displayed with interactive parallax on mobile or desktop screens, as well as on Virtual Reality devices, where viewing it also includes stereo. We present an end-to-end system for creating and viewing 3D photos, and the algorithmic and design choices therein. Our 3D photos are captured in a single shot and processed directly on a mobile device. The method starts by estimating depth from the 2D input image using a new monocular depth estimation network that is optimized for mobile devices. It performs competitively to the state-of-the-art, but has lower latency and peak memory consumption and uses an order of magnitude fewer parameters. The resulting depth is lifted to a layered depth image, and new geometry is synthesized in parallax regions. We synthesize color texture and structures in the parallax regions as well, using an inpainting network, also optimized for mobile devices, on the LDI directly. Finally, we convert the result into a mesh-based representation that can be efficiently transmitted and rendered even on low-end devices and over poor network connections. Altogether, the processing takes just a few seconds on a mobile device, and the result can be instantly viewed and shared. We perform extensive quantitative evaluation to validate our system and compare its new components against the current state-of-the-art., Comment: Project page: https://facebookresearch.github.io/one_shot_3d_photography/ Code: https://github.com/facebookresearch/one_shot_3d_photography

Video super-resolution (VSR) and frame interpolation (FI) are traditional computer vision problems, and the performance have been improving by incorporating deep learning recently. In this paper, we investigate the problem of jointly upsampling videos both in space and time, which is becoming more important with advances in display systems. One solution for this is to run VSR and FI, one by one, independently. This is highly inefficient as heavy deep neural networks (DNN) are involved in each solution. To this end, we propose an end-to-end DNN framework for the space-time video upsampling by efficiently merging VSR and FI into a joint framework. In our framework, a novel weighting scheme is proposed to fuse all input frames effectively without explicit motion compensation for efficient processing of videos. The results show better results both quantitatively and qualitatively, while reducing the computation time (\(\times \)7 faster) and the number of parameters (30%) compared to baselines. Our source code is available at https://github.com/JaeYeonKang/STVUN-Pytorch.

We present a novel 3D pose refinement approach based on differentiable rendering for objects of arbitrary categories in the wild. In contrast to previous methods, we make two main contributions: First, instead of comparing real-world images and synthetic renderings in the RGB or mask space, we compare them in a feature space optimized for 3D pose refinement. Second, we introduce a novel differentiable renderer that learns to approximate the rasterization backward pass from data instead of relying on a hand-crafted algorithm. For this purpose, we predict deep cross-domain correspondences between RGB images and 3D model renderings in the form of what we call geometric correspondence fields. These correspondence fields serve as pixel-level gradients which are analytically propagated backward through the rendering pipeline to perform a gradient-based optimization directly on the 3D pose. In this way, we precisely align 3D models to objects in RGB images which results in significantly improved 3D pose estimates. We evaluate our approach on the challenging Pix3D dataset and achieve up to 55% relative improvement compared to state-of-the-art refinement methods in multiple metrics.

When automatically generating a sentence description for an image or video, it often remains unclear how well the generated caption is grounded, that is whether the model uses the correct image regions to output particular words, or if the model is hallucinating based on priors in the dataset and/or the language model. The most common way of relating image regions with words in caption models is through an attention mechanism over the regions that are used as input to predict the next word. The model must therefore learn to predict the attentional weights without knowing the word it should localize. This is difficult to train without grounding supervision since recurrent models can propagate past information and there is no explicit signal to force the captioning model to properly ground the individual decoded words. In this work, we help the model to achieve this via a novel cyclical training regimen that forces the model to localize each word in the image after the sentence decoder generates it, and then reconstruct the sentence from the localized image region(s) to match the ground-truth. Our proposed framework only requires learning one extra fully-connected layer (the localizer), a layer that can be removed at test time. We show that our model significantly improves grounding accuracy without relying on grounding supervision or introducing extra computation during inference, for both image and video captioning tasks. Code is available at https://github.com/chihyaoma/cyclical-visual-captioning.

LiDAR point-cloud segmentation is an important problem for many applications. For large-scale point cloud segmentation, the de facto method is to project a 3D point cloud to get a 2D LiDAR image and use convolutions to process it. Despite the similarity between regular RGB and LiDAR images, we are the first to discover that the feature distribution of LiDAR images changes drastically at different image locations. Using standard convolutions to process such LiDAR images is problematic, as convolution filters pick up local features that are only active in specific regions in the image. As a result, the capacity of the network is under-utilized and the segmentation performance decreases. To fix this, we propose Spatially-Adaptive Convolution (SAC) to adopt different filters for different locations according to the input image. SAC can be computed efficiently since it can be implemented as a series of element-wise multiplications, im2col, and standard convolution. It is a general framework such that several previous methods can be seen as special cases of SAC. Using SAC, we build SqueezeSegV3 for LiDAR point-cloud segmentation and outperform all previous published methods by at least 2.0% mIoU on the SemanticKITTI benchmark. Code and pretrained model are available at https://github.com/chenfengxu714/SqueezeSegV3.

Many automated processes such as auto-piloting rely on a good semantic segmentation as a critical component. To speed up performance, it is common to downsample the input frame. However, this comes at the cost of missed small objects and reduced accuracy at semantic boundaries. To address this problem, we propose a new content-adaptive downsampling technique that learns to favor sampling locations near semantic boundaries of target classes. Cost-performance analysis shows that our method consistently outperforms the uniform sampling improving balance between accuracy and computational efficiency. Our adaptive sampling gives segmentation with better quality of boundaries and more reliable support for smaller-size objects.

This paper proposes an efficient neural network (NN) architecture design methodology called Chameleon that honors given resource constraints. Instead of developing new building blocks or using computationally-intensive reinforcement learning algorithms, our approach leverages existing efficient network building blocks and focuses on exploiting hardware traits and adapting computation resources to fit target latency and/or energy constraints. We formulate platform-aware NN architecture search in an optimization framework and propose a novel algorithm to search for optimal architectures aided by efficient accuracy and resource (latency and/or energy) predictors. At the core of our algorithm lies an accuracy predictor built atop Gaussian Process with Bayesian optimization for iterative sampling. With a one-time building cost for the predictors, our algorithm produces state-of-the-art model architectures on different platforms under given constraints in just minutes. Our results show that adapting computation resources to building blocks is critical to model performance. Without the addition of any special features, our models achieve significant accuracy improvements relative to state-of-the-art handcrafted and automatically designed architectures. We achieve 73.8% and 75.3% top-1 accuracy on ImageNet at 20ms latency on a mobile CPU and DSP. At reduced latency, our models achieve up to 8.2% (4.8%) and 6.7% (9.3%) absolute top-1 accuracy improvements compared to MobileNetV2 and MnasNet, respectively, on a mobile CPU (DSP), and 2.7% (4.6%) and 5.6% (2.6%) accuracy gains over ResNet-101 and ResNet-152, respectively, on an Nvidia GPU (Intel CPU).

In the past few years sharing photos within social networks has become very popular. In order to make these huge collections easier to explore, images are usually tagged with representative keywords such as persons, events, objects, and locations. In order to speed up the time consuming tag annotation process, tags can be propagated based on the similarity between image content and context. In this paper, we present a system for efficient geotag propagation based on a combination of object duplicate detection and user trust modeling. The geotags are propagated by training a graph based object model for each of the landmarks on a small tagged image set and finding its duplicates within a large untagged image set. Based on the established correspondences between these two image sets and the reliability of the user, tags are propagated from the tagged to the untagged images. The user trust modeling reduces the risk of propagating wrong tags caused by spamming or faulty annotation. The effectiveness of the proposed method is demonstrated through a set of experiments on an image database containing various landmarks.

We propose a novel value-aware quantization which applies aggressively reduced precision to the majority of data while separately handling a small amount of large values in high precision, which reduces total quantization errors under very low precision. We present new techniques to apply the proposed quantization to training and inference. The experiments show that our method with 3-bit activations (with 2% of large ones) can give the same training accuracy as full-precision one while offering significant (41.6% and 53.7%) reductions in the memory cost of activations in ResNet-152 and Inception-v3 compared with the state-of-the-art method. Our experiments also show that deep networks such as Inception-v3, ResNet-101 and DenseNet-121 can be quantized for inference with 4-bit weights and activations (with 1% 16-bit data) within 1% top-1 accuracy drop.

Designing accurate and efficient ConvNets for mobile devices is challenging because the design space is combinatorially large. Due to this, previous neural architecture search (NAS) methods are computationally expensive. ConvNet architecture optimality depends on factors such as input resolution and target devices. However, existing approaches are too resource demanding for case-by-case redesigns. Also, previous work focuses primarily on reducing FLOPs, but FLOP count does not always reflect actual latency. To address these, we propose a differentiable neural architecture search (DNAS) framework that uses gradient-based methods to optimize ConvNet architectures, avoiding enumerating and training individual architectures separately as in previous methods. FBNets (Facebook-Berkeley-Nets), a family of models discovered by DNAS surpass state-of-the-art models both designed manually and generated automatically. FBNet-B achieves 74.1% top-1 accuracy on ImageNet with 295M FLOPs and 23.1 ms latency on a Samsung S8 phone, 2.4x smaller and 1.5x faster than MobileNetV2-1.3 with similar accuracy. Despite higher accuracy and lower latency than MnasNet, we estimate FBNet-B's search cost is 420x smaller than MnasNet's, at only 216 GPU-hours. Searched for different resolutions and channel sizes, FBNets achieve 1.5% to 6.4% higher accuracy than MobileNetV2. The smallest FBNet achieves 50.2% accuracy and 2.9 ms latency (345 frames per second) on a Samsung S8. Over a Samsung-optimized FBNet, the iPhone-X-optimized model achieves a 1.4x speedup on an iPhone X. FBNet models are open-sourced at https://github. com/facebookresearch/mobile-vision.

Tagging in online social networks is very popular these days, as it facilitates search and retrieval of multimedia content. However, noisy and spam annotations often make it difficult to perform an efficient search. Users may make mistakes in tagging and irrelevant tags and content may be maliciously added for advertisement or self-promotion. This article surveys recent advances in techniques for combatting such noise and spam in social tagging. We classify the state-of-the-art approaches into a few categories and study representative examples in each. We also qualitatively compare and contrast them and outline open issues for future research.

We demonstrate EigenNews, a personalized television news system. Upon visiting the EigenNews website, a user is shown a variety of news videos which have been automatically selected based on her individual preferences. These videos are extracted from 16 continually recorded television programs using a multimodal segmentation algorithm. Relevant metadata for each video are generated by linking videos to online news articles. Selected news videos can be watched in three different layouts and on various devices.

Next-generation news video consumption will be more personalized, device agnostic, and pooled from many different sources. The EigenNews system provides each viewer with a personalized newscast filled with stories that matter most to them. While personalized text-based news services already exist, personalized video news streams are still a critically missing technology. Our system records multiple channels, segments each news stream into individual stories, discovers links between the stories themselves and links between the stories and online articles, and generates a personalized playlist for each user. The dynamic mixing and aggregation of broadcast news videos from multiple sources greatly enriches the news watching experience by providing more comprehensive coverage and varying perspectives. Over time, the website collects click and view histories and uses this information to continuously optimize the personalization.

Many large collections of news videos dating back several decades can now be accessed online. For users to easily retrieve a compilation of stories on a particular event/topic and to quickly sample each story clip, all the news videos must be precisely segmented into stories and a representative video summary must be generated for each story. In this paper, we demonstrate that effectively exploiting the visual similarities pervasive in all news videos can greatly help to fulfill these technical requirements and thus enable the dynamic retrieval and mixing of small news video fragments. Two new algorithms are developed to accurately detect two important sources of visual similarity: (1) similar preview and story frames, and (2) repeated appearances of a news anchor. As a result, valuable sources of preview clips and informative clues about story boundaries are obtained from identification of these visual similarities. The retrieval engine implemented in both algorithms employs compact global image signatures and requires a small memory footprint, so that many instances of the detection algorithms can run concurrently on the same server for fast processing of a large collection of news videos. At the same time, the retrieval engine is robust to the large appearance variations encountered in the preview matching and anchor detection problems. In addition, since the video frame's color information is not required in our algorithms, both modern color and vintage black-and-white news footage can be processed in the same framework.

The amount of information that people share on social networks is constantly increasing. People also comment, annotate, and tag their own content (videos, photos, notes, etc.), as well as the content of others. In many cases, the content is tagged manually. One way to make this time-consuming manual tagging process more efficient is to propagate tags from a small set of tagged images to the larger set of untagged images automatically. In such a scenario, however, a wrong or a spam tag can damage the integrity and reliability of the automated propagation system. Users may make mistakes in tagging, or irrelevant tags and content may be added maliciously for advertisement or self-promotion. Therefore, a certain mechanism insuring the trustworthiness of users or published content is needed. In this chapter, we discuss several image retrieval methods based on tags, various approaches to trust modeling and spam protection in social networks, and trust modeling in geotagging systems. We then consider a specific example of automated geotag propagation system that adopts a user trust model. The tag propagation in images relies on the similarity between image content (famous landmarks) and its context (associated geotags). For each tagged image, similar untagged images are found by the robust graph-based object duplicate detection, and the known tags are propagated accordingly. The user trust value is estimated based on a social feedback from the users of the photo-sharing system, and only tags from trusted users are propagated. This approach demonstrates that a practical tagging system significantly benefits from the intelligent combination of efficient propagation algorithm and a user-centered trust model.

In this paper, the authors analyze their graph-based approach for 2D and 3D object duplicate detection in still images. A graph model is used to represent the 3D spatial information of the object based on the features extracted from training images to avoid explicit and complex 3D object modeling. Therefore, improved performance can be achieved in comparison to existing methods in terms of both robustness and computational complexity. Different limitations of this approach are analyzed by evaluating performance with respect to the number of training images and calculation of optimal parameters in a number of applications. Furthermore, effectiveness of object duplicate detection algorithm is measured over different object classes. The authors’ method is shown to be robust in detecting the same objects even when images with objects are taken from different viewpoints or distances.

With the rapid growth of digital photography, sharing of photos with friends and family has become very popular. When people share their photos, they usually organize them into albums according to events or places. To tell the story of some important events in one’s life, it is desirable to have an efficient summarization tool which can help people to receive a quick overview of an album containing large number of photos. In this paper, we present and analyze an approach for photo album summarization through a novel social game “Epitome” as a Facebook application. This social game can collect research data, and, at the same time, it provides a collage or a cover photo of the user’s photo album, while the user enjoys playing the game. The proof of concept of the proposed method is demonstrated through a set of experiments on several photo albums. As a benchmark comparison to this game, we perform automatic visual analysis considering several state-of-the-art features. We also evaluate the usability of the game by making use of a questionnaire on several subjects who played the “Epitome” game. Furthermore, we address privacy issues concerning shared photos in Facebook applications.

In this paper, we propose an approach for photo album summarization through a novel social game "Epitome" for mobile phones. Our approach to album summarization consists of two games: "Select the Best!" and "Split it!". The goal of the first game is to allow a user to select the most representative photo of a reduced set of images, while in the second game, the user has to split the reduced set into two distinct parts. As it could be time-consuming to look at a huge collection of photos on a mobile phone, it is more enjoyable and pleasant to show only a limited number of images which can be fit into one mobile screen. The results obtained in these games are combined to produce a summarization and are then compared with the results of other users. As a final result, a unique summarization sequence of photos is determined. The determined sequence of photos can be used to create a collage of one album or a cover for an album. The proof of concept of the proposed method is demonstrated through a set of experiments on several photo collections.

In this paper, we consider the use of object duplicate detection for the propagation of geotags from a small set of images with location names (IPTC) to a large set of non-tagged images. The motivation behind this idea is that images of individual locations usually contain specific objects such as monuments, buildings or signs. Therefore, object duplicate detection can be used to establish the correspondence between tagged and non-tagged images. Our recent graph based object duplicate detection approach is adapted for this task. The effectiveness of the approach is demonstrated through a set of experiments considering various locations.

Over the last few years, social network systems have greatly increased users’ involvement in online content creation and annotation. Since such systems usually need to deal with a large amount of multimedia data, it becomes desirable to realize an interactive service that minimizes tedious and time consuming manual annotation. In this paper, we propose an interactive online platform that is capable of performing semi-automatic image annotation and tag recommendation for an extensive online database. First, when the user marks a specific object in an image, the system performs an object duplicate detection and returns the search results with images containing similar objects. Then, the annotation of the object can be performed in two ways: (1) In the tag recommendation process, the system recommends tags associated with the object in images of the search results, among which, the user can accept some tags for the object in the given image. (2) In the tag propagation process, when the user enters his/her tag for the object, it is propagated to images in the search results. Different techniques to speed-up the process of indexing and retrieval are presented in this paper and their effectiveness demonstrated through a set of experiments considering various classes of objects.

In this paper, we consider the challenging problem of object duplicate detection and localization. Several applications require efficient object duplicate detection methods, such as automatic video and image tagging, video surveillance, and high level image or video search. In this paper, a novel graph-based approach for 3D object duplicate detection in still images is proposed. A graph model is used to represent the spatial information of the object in order to avoid making an explicit 3D object model. Therefore, better performance is achieved in terms of robustness and computational complexity.

Several applications require accurate and efficient object duplicate detection methods, such as automatic video and image tag propagation, video surveillance, and high level image or video search. In this paper, we explore the limits of our recently proposed graph-based object duplicate detection method. The dependency of the performance with respect to the number of training images is assessed and the optimal detection parameters are determined. Furthermore, the differences among various object classes are analyzed. In this way, this paper provides an in-depth analysis of the graph based object duplicate detection method.

Many images uploaded to social networks are related to travel, since people consider traveling to be an important event in their life. However, a significant amount of travel images on the Internet lack proper geographical annotations or tags. In many cases, the images are tagged manually. One way to make this time-consuming manual tagging process more efficient is to propagate tags from a small set of tagged images to the larger set of untagged images automatically. In this paper, we present a system for automatic geotag propagation in images based on the similarity between image content (famous landmarks) and its context (associated geotags). In such a scenario, however, an incorrect or a spam tag can damage the integrity and reliability of the automated propagation system. Therefore, for reliable geotags propagation, we suggest adopting a user trust model based on social feedback from the users of the photo-sharing system. We compare this socially-driven approach with other user trust models via experiments and subjective testing on an image database of various famous landmarks. Results demonstrate that relying on user feedback is more efficient, since the number of propagated tags more than doubles without loss of accuracy compared to using other models or propagating without trust modeling.

With the rapid growth of digital photography, sharing of photos with friends and family has become very popular. When people share their photos, they usually organize them in albums according to events or places. To tell the story of some important events in one's life, it is desirable to have an efficient summarization tool which can help people to get a quick overview of an album containing huge number of photos. In this paper, we analyze our approach for photo album summarization through a novel social game ``Epitome'' as a Facebook application. Our social game can collect research data and, at the same time, it provides a collage or a cover photo of the user's photo album, while, at the same time, the user enjoys playing the game. As a benchmark comparison to this game, we performed automatic visual analysis considering several state-of-the-art features.

Spatial region (image) segmentation is a fundamental step for many computer vision applications. Although many methods have been proposed, less work has been done in developing suitable evaluation methodologies for comparing different approaches. The main problem of general purpose segmentation evaluation is the dilemma between objectivity and generality. Recently, figure ground segmentation evaluation has been proposed to solve this problem by defining an unambiguous ground truth using the most salient foreground object. Although the annotation of a single foreground object is less complex than the annotation of all regions within an image, it is still quite time consuming, especially for videos. A novel framework incorporating background subtraction for automatic ground truth generation and different foreground evaluation measures is proposed, that allows to effectively and efficiently evaluate the performance of image segmentation approaches. The experiments show that the objective measures are comparable to the subjective assessment and that there is only a slight difference between manually annotated and automatically generated ground truth.