Current dataset collection methods typically scrape large amounts of data from the web. While this technique is extremely scalable, data collected in this way tends to reinforce stereotypical biases, can contain personally identifiable information, and typically originates from Europe and North America. In this work, we rethink the dataset collection paradigm and introduce GeoDE, a geographically diverse dataset with 61,940 images from 40 classes and 6 world regions, and no personally identifiable information, collected by soliciting images from people across the world. We analyse GeoDE to understand differences in images collected in this manner compared to web-scraping. Despite the smaller size of this dataset, we demonstrate its use as both an evaluation and training dataset, allowing us to highlight shortcomings in current models, as well as demonstrate improved performance even when training on this small dataset. We release the full dataset and code at https://geodiverse-data-collection.cs.princeton.edu/

Perception-based image analysis technologies can be used to help visually impaired people take better quality pictures by providing automated guidance, thereby empowering them to interact more confidently on social media. The photographs taken by visually impaired users often suffer from one or both of two kinds of quality issues: technical quality (distortions), and semantic quality, such as framing and aesthetic composition. Here we develop tools to help them minimize occurrences of common technical distortions, such as blur, poor exposure, and noise. We do not address the complementary problems of semantic quality, leaving that aspect for future work. The problem of assessing, and providing actionable feedback on the technical quality of pictures captured by visually impaired users is hard enough, owing to the severe, commingled distortions that often occur. To advance progress on the problem of analyzing and measuring the technical quality of visually impaired user-generated content (VI-UGC), we built a very large and unique subjective image quality and distortion dataset. This new perceptual resource, which we call the LIVE-Meta VI-UGC Database, contains 40K real-world distorted VI-UGC images and 40K patches, on which we recorded 2.7M human perceptual quality judgments and 2.7M distortion labels. Using this psychometric resource we also created an automatic limited vision picture quality and distortion predictor that learns local-to-global spatial quality relationships, achieving state-of-the-art prediction performance on VI-UGC pictures, significantly outperforming existing picture quality models on this unique class of distorted picture data. We also created a prototype feedback system that helps to guide users to mitigate quality issues and take better quality pictures, by creating a multi-task learning framework. The dataset and models can be accessed at: https://github.com/mandal-cv/visimpaired.

A visual counterfactual explanation replaces image regions in a query image with regions from a distractor image such that the system’s decision on the transformed image changes to the distractor class. In this work, we present a novel framework for computing visual counterfactual explanations based on two key ideas. First, we enforce that the replaced and replacer regions contain the same semantic part, resulting in more semantically consistent explanations. Second, we use multiple distractor images in a computationally efficient way and obtain more discriminative explanations with fewer region replacements. Our approach is $$\mathbf {27\%}$$ more semantically consistent and an order of magnitude faster than a competing method on three fine-grained image recognition datasets. We highlight the utility of our counterfactuals over existing works through machine teaching experiments where we teach humans to classify different bird species. We also complement our explanations with the vocabulary of parts and attributes that contributed the most to the system’s decision. In this task as well, we obtain state-of-the-art results when using our counterfactual explanations relative to existing works, reinforcing the importance of semantically consistent explanations. Source code is available at github.com/facebookresearch/visual-counterfactuals .

One of the factors that have hindered progress in the areas of sign language recognition, translation, and production is the absence of large annotated datasets. Towards this end, we introduce How2Sign, a multimodal and multiview continuous American Sign Language (ASL) dataset, consisting of a parallel corpus of more than 80 hours of sign language videos and a set of corresponding modalities including speech, English transcripts, and depth. A three-hour subset was further recorded in the Panoptic studio enabling detailed 3D pose estimation. To evaluate the potential of How2Sign for real-world impact, we conduct a study with ASL signers and show that synthesized videos using our dataset can indeed be understood. The study further gives insights on challenges that computer vision should address in order to make progress in this field.

No-reference (NR) perceptual video quality assessment (VQA) is a complex, unsolved, and important problem for social and streaming media applications. Efficient and accurate video quality predictors are needed to monitor and guide the processing of billions of shared, often imperfect, user-generated content (UGC). Unfortunately, current NR models are limited in their prediction capabilities on real-world, "in-the-wild" UGC video data. To advance progress on this problem, we created the largest (by far) subjective video quality dataset, containing 38,811 real-world distorted videos and 116,433 space-time localized video patches (‘v-patches’), and 5.5M human perceptual quality annotations. Using this, we created two unique NR-VQA models: (a) a local-to-global region-based NR VQA architecture (called PVQ) that learns to predict global video quality and achieves state-of-the-art performance on 3 UGC datasets, and (b) a first-of-a-kind space-time video quality mapping engine (called PVQ Mapper) that helps localize and visualize perceptual distortions in space and time. The entire dataset and prediction models are freely available at https://live.ece.utexas.edu/research.php.

This paper presents a novel task together with a new benchmark for detecting generic, taxonomy-free event boundaries that segment a whole video into chunks. Conventional work in temporal video segmentation and action detection focuses on localizing pre-defined action categories and thus does not scale to generic videos. Cognitive Science has known since last century that humans consistently segment videos into meaningful temporal chunks. This segmentation happens naturally, without pre-defined event categories and without being explicitly asked to do so. Here, we repeat these cognitive experiments on mainstream CV datasets; with our novel annotation guideline which addresses the complexities of taxonomy-free event boundary annotation, we introduce the task of Generic Event Boundary Detection (GEBD) and the new benchmark Kinetics-GEBD. We view GEBD as an important stepping stone towards understanding the video as a whole, and believe it has been previously neglected due to a lack of proper task definition and annotations. Through experiment and human study we demonstrate the value of the annotations. Further, we benchmark supervised and un-supervised GEBD approaches on the TAPOS dataset and our Kinetics-GEBD. We release our annotations and baseline codes at CVPR’21 LOVEU Challenge: https://sites.google.com/view/loveucvpr21.

Existing models often leverage co-occurrences between objects and their context to improve recognition accuracy. However, strongly relying on context risks a model's generalizability, especially when typical co-occurrence patterns are absent. This work focuses on addressing such contextual biases to improve the robustness of the learnt feature representations. Our goal is to accurately recognize a category in the absence of its context, without compromising on performance when it co-occurs with context. Our key idea is to decorrelate feature representations of a category from its co-occurring context. We achieve this by learning a feature subspace that explicitly represents categories occurring in the absence of context along side a joint feature subspace that represents both categories and context. Our very simple yet effective method is extensible to two multi-label tasks -- object and attribute classification. On 4 challenging datasets, we demonstrate the effectiveness of our method in reducing contextual bias.

Pre-training convolutional neural networks with weakly-supervised and self-supervised strategies is becoming increasingly popular for several computer vision tasks. However, due to the lack of strong discriminative signals, these learned representations may overfit to the pre-training objective (e.g., hashtag prediction) and not generalize well to downstream tasks. In this work, we present a simple strategy - ClusterFit to improve the robustness of the visual representations learned during pre-training. Given a dataset, we (a) cluster its features extracted from a pre-trained network using k-means and (b) re-train a new network from scratch on this dataset using cluster assignments as pseudo-labels. We empirically show that clustering helps reduce the pre-training task-specific information from the extracted features thereby minimizing overfitting to the same. Our approach is extensible to different pre-training frameworks -- weak- and self-supervised, modalities -- images and videos, and pre-training tasks -- object and action classification. Through extensive transfer learning experiments on 11 different target datasets of varied vocabularies and granularities, we show that ClusterFit significantly improves the representation quality compared to the state-of-the-art large-scale (millions / billions) weakly-supervised image and video models and self-supervised image models.

Blind or no-reference (NR) perceptual picture quality prediction is a difficult, unsolved problem of great consequence to the social and streaming media industries that impacts billions of viewers daily. Unfortunately, popular NR prediction models perform poorly on real-world distorted pictures. To advance progress on this problem, we introduce the largest (by far) subjective picture quality database, containing about 40, 000 real-world distorted pictures and 120, 000 patches, on which we collected about 4M human judgments of picture quality. Using these picture and patch quality labels, we built deep region-based architectures that learn to produce state-of-the-art global picture quality predictions as well as useful local picture quality maps. Our innovations include picture quality prediction architectures that produce global-to-local inferences as well as local-to-global inferences (via feedback). The dataset and source code are available at https: //live.ece.utexas.edu/research.php.

Current fully-supervised video datasets consist of only a few hundred thousand videos and fewer than a thousand domain-specific labels. This hinders the progress towards advanced video architectures. This paper presents an in-depth study of using large volumes of web videos for pre-training video models for the task of action recognition. Our primary empirical finding is that pre-training at a very large scale (over 65 million videos), despite on noisy social-media videos and hashtags, substantially improves the state-of-the-art on three challenging public action recognition datasets. Further, we examine three questions in the construction of weakly-supervised video action datasets. First, given that actions involve interactions with objects, how should one construct a verb-object pre-training label space to benefit transfer learning the most? Second, frame-based models perform quite well on action recognition; is pre-training for good image features sufficient or is pre-training for spatio-temporal features valuable for optimal transfer learning? Finally, actions are generally less well-localized in long videos vs. short videos; since action labels are provided at a video level, how should one choose video clips for best performance, given some fixed budget of number or minutes of videos?

Highlight detection has the potential to significantly ease video browsing, but existing methods often suffer from expensive supervision requirements, where human viewers must manually identify highlights in training videos. We propose a scalable unsupervised solution that exploits video duration as an implicit supervision signal. Our key insight is that video segments from shorter user-generated videos are more likely to be highlights than those from longer videos, since users tend to be more selective about the content when capturing shorter videos. Leveraging this insight, we introduce a novel ranking framework that prefers segments from shorter videos, while properly accounting for the inherent noise in the (unlabeled) training data. We use it to train a highlight detector with 10M hashtagged Instagram videos. In experiments on two challenging public video highlight detection benchmarks, our method substantially improves the state-of-the-art for unsupervised highlight detection.

Weakly supervised object detection aims at reducing the amount of supervision required to train detection models. Such models are traditionally learned from images/videos labelled only with the object class and not the object bounding box. In our work, we try to leverage not only the object class labels but also the action labels associated with the data. We show that the action depicted in the image/video can provide strong cues about the location of the associated object. We learn a spatial prior for the object dependent on the action (e.g. "ball" is closer to "leg of the person" in "kicking ball"), and incorporate this prior to simultaneously train a joint object detection and action classification model. We conducted experiments on both video datasets and image datasets to evaluate the performance of our weakly supervised object detection model. Our approach outperformed the current state-of-the-art (SOTA) method by more than 6% in mAP on the Charades video dataset.

Over-the-top mobile adaptive video streaming is invariably influenced by volatile network conditions, which can cause playback interruptions (stalling or rebuffering events) and bitrate fluctuations, thereby impairing users' quality of experience (QoE). Video quality assessment models that can accurately predict users' QoE under such volatile network conditions are rapidly gaining attention, since these methods could enable more efficient design of quality control protocols for media-driven services such as YouTube, Amazon, Netflix, and many others. However, the development of improved QoE prediction models requires data sets of videos afflicted with diverse stalling events that have been labeled with ground-truth subjective opinion scores. Toward this end, we have created a new mobile video quality database that we call LIVE Mobile Stall Video Database-II. Our database contains a total of 174 videos afflicted with distortions caused by 26 different stalling patterns. We describe the way we simulated the diverse stalling events to create a corpus of distorted videos, and we detail the human study we conducted to obtain continuous-time subjective scores from 54 subjects. We also present the outcomes of our comprehensive analysis of the impact of several factors that influence subjective QoE, and report the performance of existing QoE-prediction models on our data set. We are making the database (videos, subjective data, and video metadata) publicly available in order to help the advance state-of-the-art research on user-centric mobile network planning and management. The database may be accessed at http://live.ece.utexas.edu/research/LIVEStallStudy/liveMobile.html.

Digital videos often contain visual distortions that are introduced by the camera's hardware or processing software during the capture process. These distortions often detract from a viewer's quality of experience. Understanding how human observers perceive the visual quality of digital videos is of great importance to camera designers. Thus, the development of automatic objective methods that accurately quantify the impact of visual distortions on perception has greatly accelerated. Video quality algorithm design and verification require realistic databases of distorted videos and human judgments of them. However, most current publicly available video quality databases have been created under highly controlled conditions using graded, simulated, and post-capture distortions (such as jitter and compression artifacts) on high-quality videos. The commercial plethora of hand-held mobile video capture devices produces videos often afflicted by a variety of complex distortions generated during the capturing process. These in-capture distortions are not well-modeled by the synthetic, post-capture distortions found in existing VQA databases. Toward overcoming this limitation, we designed and created a new database that we call the LIVE-Qualcomm mobile in-capture video quality database, comprising a total of 208 videos, which model six common in-capture distortions. We also conducted a subjective quality assessment study using this database, in which each video was assessed by 39 unique subjects. Furthermore, we evaluated several top-performing no-reference IQA and VQA algorithms on the new database and studied how real-world in-capture distortions challenge both human viewers as well as automatic perceptual quality prediction models. The new database is freely available at: http://live.ece.utexas.edu/research/incaptureDatabase/index.html.

Over-the-top adaptive video streaming services are frequently impacted by fluctuating network conditions that can lead to rebuffering events (stalling events) and sudden bitrate changes. These events visually impact video consumers' quality of experience (QoE) and can lead to consumer churn. The development of models that can accurately predict viewers' instantaneous subjective QoE under such volatile network conditions could potentially enable the more efficient design of quality-control protocols for media-driven services, such as YouTube, Amazon, Netflix, and so on. However, most existing models only predict a single overall QoE score on a given video and are based on simple global video features, without accounting for relevant aspects of human perception and behavior. We have created a QoE evaluator, called the time-varying QoE Indexer, that accounts for interactions between stalling events, analyzes the spatial and temporal content of a video, predicts the perceptual video quality, models the state of the client-side data buffer, and consequently predicts continuous-time quality scores that agree quite well with human opinion scores. The new QoE predictor also embeds the impact of relevant human cognitive factors, such as memory and recency, and their complex interactions with the video content being viewed. We evaluated the proposed model on three different video databases and attained standout QoE prediction performance.

We designed and created a new video database that models a variety of complex distortions generated during the video capturing process on hand-held mobile capturing devices. We describe the content and characteristics of the new database, which we call the LIVE Mobile In-Capture Video Quality Database. It comprises a total of 208 videos that were captured using eight different smart-phones and were affected by six common in-capture distortions. We also conducted a subjective video quality assessment study using this data, wherein each video was assessed by 36 unique subjects. We evaluated several top-performing No-Reference IQA and VQA algorithms on the new database and find insights on how real-world in-capture distortions challenge both human subjects as well as automatic perceptual quality prediction models.

No-Reference (NR) video quality assessment (VQA) models are gaining popularity as they offer scope for broader applicability to user-uploaded video-centric services such as YouTube and Facebook, where the pristine references are unavailable. However, there are few, well-performing NR-VQA models owing to the difficulty of the problem. We propose a novel NR video quality predictor that solely relies on the `quality-aware' natural statistical models in the space-time domain. The proposed quality predictor called Self-reference based LEarning-free Evaluator of Quality (SLEEQ) consists of three components: feature extraction in the spatial and temporal domains, motion-based feature fusion, and spatial-temporal feature pooling to derive a single quality score for a given video. SLEEQ achieves higher than 0.9 correlation with the subjective video quality scores on tested public databases and thus outperforms the existing NR VQA models.

Being able to automatically predict digital picture quality, as perceived by human observers, has become important in many applications where humans are the ultimate consumers of displayed visual information. Standard dynamic range (SDR) images provide 8 b/color/pixel. High dynamic range (HDR) images, which are usually created from multiple exposures of the same scene, can provide 16 or 32 b/color/pixel, but must be tonemapped to SDR for display on standard monitors. Multi-exposure fusion techniques bypass HDR creation, by fusing the exposure stack directly to SDR format while aiming for aesthetically pleasing luminance and color distributions. Here, we describe a new no-reference image quality assessment (NR IQA) model for HDR pictures that is based on standard measurements of the bandpass and on newly conceived differential natural scene statistics (NSS) of HDR pictures. We derive an algorithm from the model which we call the HDR IMAGE GRADient-based Evaluator. NSS models have previously been used to devise NR IQA models that effectively predict the subjective quality of SDR images, but they perform significantly worse on tonemapped HDR content. Toward ameliorating this we make here the following contributions: 1) we design HDR picture NR IQA models and algorithms using both standard space-domain NSS features as well as novel HDR-specific gradient-based features that significantly elevate prediction performance; 2) we validate the proposed models on a large-scale crowdsourced HDR image database; and 3) we demonstrate that the proposed models also perform well on legacy natural SDR images. The software is available at: http://live.ece.utexas.edu/research/Quality/higradeRelease.zip.

Measuring digital picture quality, as perceived by human observers, is increasingly important in many applications in which humans are the ultimate consumers of visual information. Standard dynamic range (SDR) images provide 8 b/color/pixel. High dynamic range (HDR) images, usually created from multiple exposures of the same scene, can provide 16 or 32 b/color/pixel, but need to be tonemapped to SDR for display on standard monitors. Multiexposure fusion (MEF) techniques bypass HDR creation by fusing an exposure stack directly to SDR images to achieve aesthetically pleasing luminance and color distributions. Many HDR and MEF databases have a relatively small number of images and human opinion scores, obtained under stringently controlled conditions, thereby limiting realistic viewing. Moreover, many of these databases are intended to compare tone-mapping algorithms, rather than being specialized for developing and comparing image quality assessment models. To overcome these challenges, we conducted a massively crowdsourced online subjective study. The primary contributions described in this paper are: 1) the new ESPL-LIVE HDR Image Database that we created containing diverse images obtained by tone-mapping operators and MEF algorithms, with and without post-processing; 2) a large-scale subjective study that we conducted using a crowdsourced platform to gather more than 300 000 opinion scores on 1811 images from over 5000 unique observers; and 3) a detailed study of the correlation performance of the state-of-the-art no-reference image quality assessment algorithms against human opinion scores of these images. The database is available at http://signal.ece.utexas.edu/%7Edebarati/HDRDatabase.zip.

Most publicly available image quality databases have been created under highly controlled conditions by introducing graded simulated distortions onto high-quality photographs. However, images captured using typical real-world mobile camera devices are usually afflicted by complex mixtures of multiple distortions, which are not necessarily well-modeled by the synthetic distortions found in existing databases. The originators of existing legacy databases usually conducted human psychometric studies to obtain statistically meaningful sets of human opinion scores on images in a stringently controlled visual environment, resulting in small data collections relative to other kinds of image analysis databases. Toward overcoming these limitations, we designed and created a new database that we call the LIVE In the Wild Image Quality Challenge Database, which contains widely diverse authentic image distortions on a large number of images captured using a representative variety of modern mobile devices. We also designed and implemented a new online crowdsourcing system, which we have used to conduct a very large-scale, multi-month image quality assessment (IQA) subjective study. Our database consists of over 350 000 opinion scores on 1162 images evaluated by over 8100 unique human observers. Despite the lack of control over the experimental environments of the numerous study participants, we demonstrate excellent internal consistency of the subjective data set. We also evaluate several top-performing blind IQA algorithms on it and present insights on how the mixtures of distortions challenge both end users as well as automatic perceptual quality prediction models. The new database is available for public use at http://live.ece.utexas.edu/research/ChallengeDB/index.html.

Current top-performing blind image quality assessment (IQA) models rely on benchmark databases comprising of singly distorted images, thereby learning image features that are only adequate to predict human perceived visual quality on such inauthentic distortions. Furthermore, the underlying image features of these models are often extracted from the achromatic luminance channel and could sometimes fail to account for the loss of their perceived quality that might potentially be distinctly captured in a different image modality. In this work, we propose a novel IQA model that focuses on the natural scene statistics of images afflicted with complex mixtures of unknown, authentic distortions. We derive several feature maps in different perceptually relevant color spaces and extract a large number of image features from them. We demonstrate the remarkable competence of our features in improving the automatic perceptual quality prediction on images containing both synthetic and authentic distortions.

The vast majority of today's internet video services are consumed over-the-top (OTT) via reliable streaming (HTTP via TCP), where the primary noticeable delivery-related impairments are startup delay and stalling. In this paper we introduce an objective model called the delivery quality score (DQS) model, to predict user's QoE in the presence of such impairments. We describe a large subjective study that we carried out to tune and validate this model. Our experiments demonstrate that the DQS model correlates highly with the subjective data and that it outperforms other emerging models.