The MS COCO (Microsoft Common Objects in Context) dataset is a large-scale object detection, segmentation, key-point detection, and captioning dataset. The dataset consists of 328K images.

10,098 PAPERS • 92 BENCHMARKS

The Densely Annotation Video Segmentation dataset (DAVIS) is a high quality and high resolution densely annotated video segmentation dataset under two resolutions, 480p and 1080p. There are 50 video sequences with 3455 densely annotated frames in pixel level. 30 videos with 2079 frames are for training and 20 videos with 1376 frames are for validation.

628 PAPERS • 13 BENCHMARKS

MSR-VTT (Microsoft Research Video to Text) is a large-scale dataset for the open domain video captioning, which consists of 10,000 video clips from 20 categories, and each video clip is annotated with 20 English sentences by Amazon Mechanical Turks. There are about 29,000 unique words in all captions. The standard splits uses 6,513 clips for training, 497 clips for validation, and 2,990 clips for testing.

520 PAPERS • 7 BENCHMARKS

The 3D Poses in the Wild dataset is the first dataset in the wild with accurate 3D poses for evaluation. While other datasets outdoors exist, they are all restricted to a small recording volume. 3DPW is the first one that includes video footage taken from a moving phone camera.

338 PAPERS • 5 BENCHMARKS

The Microsoft Research Video Description Corpus (MSVD) dataset consists of about 120K sentences collected during the summer of 2010. Workers on Mechanical Turk were paid to watch a short video snippet and then summarize the action in a single sentence. The result is a set of roughly parallel descriptions of more than 2,000 video snippets. Because the workers were urged to complete the task in the language of their choice, both paraphrase and bilingual alternations are captured in the data.

278 PAPERS • 3 BENCHMARKS

The Extended Cohn-Kanade (CK+) dataset contains 593 video sequences from a total of 123 different subjects, ranging from 18 to 50 years of age with a variety of genders and heritage. Each video shows a facial shift from the neutral expression to a targeted peak expression, recorded at 30 frames per second (FPS) with a resolution of either 640x490 or 640x480 pixels. Out of these videos, 327 are labelled with one of seven expression classes: anger, contempt, disgust, fear, happiness, sadness, and surprise. The CK+ database is widely regarded as the most extensively used laboratory-controlled facial expression classification database available, and is used in the majority of facial expression classification methods.

220 PAPERS • 2 BENCHMARKS

Multimodal EmotionLines Dataset (MELD) has been created by enhancing and extending EmotionLines dataset. MELD contains the same dialogue instances available in EmotionLines, but it also encompasses audio and visual modality along with text. MELD has more than 1400 dialogues and 13000 utterances from Friends TV series. Multiple speakers participated in the dialogues. Each utterance in a dialogue has been labeled by any of these seven emotions -- Anger, Disgust, Sadness, Joy, Neutral, Surprise and Fear. MELD also has sentiment (positive, negative and neutral) annotation for each utterance.

208 PAPERS • 2 BENCHMARKS

The GOT-10k dataset contains more than 10,000 video segments of real-world moving objects and over 1.5 million manually labelled bounding boxes. The dataset contains more than 560 classes of real-world moving objects and 80+ classes of motion patterns.

202 PAPERS • 2 BENCHMARKS

The ShanghaiTech Campus dataset has 13 scenes with complex light conditions and camera angles. It contains 130 abnormal events and over 270, 000 training frames. Moreover, both the frame-level and pixel-level ground truth of abnormal events are annotated in this dataset.

162 PAPERS • 4 BENCHMARKS

YouCook2 is the largest task-oriented, instructional video dataset in the vision community. It contains 2000 long untrimmed videos from 89 cooking recipes; on average, each distinct recipe has 22 videos. The procedure steps for each video are annotated with temporal boundaries and described by imperative English sentences (see the example below). The videos were downloaded from YouTube and are all in the third-person viewpoint. All the videos are unconstrained and can be performed by individual persons at their houses with unfixed cameras. YouCook2 contains rich recipe types and various cooking styles from all over the world.

158 PAPERS • 7 BENCHMARKS

This paper introduces the pipeline to scale the largest dataset in egocentric vision EPIC-KITCHENS. The effort culminates in EPIC-KITCHENS-100, a collection of 100 hours, 20M frames, 90K actions in 700 variable-length videos, capturing long-term unscripted activities in 45 environments, using head-mounted cameras. Compared to its previous version (EPIC-KITCHENS-55), EPIC-KITCHENS-100 has been annotated using a novel pipeline that allows denser (54% more actions per minute) and more complete annotations of fine-grained actions (+128% more action segments). This collection also enables evaluating the "test of time" - i.e. whether models trained on data collected in 2018 can generalise to new footage collected under the same hypotheses albeit "two years on". The dataset is aligned with 6 challenges: action recognition (full and weak supervision), action detection, action anticipation, cross-modal retrieval (from captions), as well as unsupervised domain adaptation for action recognition.

133 PAPERS • 7 BENCHMARKS

The SumMe dataset is a video summarization dataset consisting of 25 videos, each annotated with at least 15 human summaries (390 in total).

124 PAPERS • 3 BENCHMARKS

The TVQA dataset is a large-scale video dataset for video question answering. It is based on 6 popular TV shows (Friends, The Big Bang Theory, How I Met Your Mother, House M.D., Grey's Anatomy, Castle). It includes 152,545 QA pairs from 21,793 TV show clips. The QA pairs are split into the ratio of 8:1:1 for training, validation, and test sets. The TVQA dataset provides the sequence of video frames extracted at 3 FPS, the corresponding subtitles with the video clips, and the query consisting of a question and four answer candidates. Among the four answer candidates, there is only one correct answer.

116 PAPERS • 3 BENCHMARKS

Cholec80 is an endoscopic video dataset containing 80 videos of cholecystectomy surgeries performed by 13 surgeons. The videos are captured at 25 fps and downsampled to 1 fps for processing. The whole dataset is labeled with the phase and tool presence annotations. The phases have been defined by a senior surgeon in Strasbourg hospital, France. Since the tools are sometimes hardly visible in the images and thus difficult to be recognized visually, a tool is defined as present in an image if at least half of the tool tip is visible.

96 PAPERS • 2 BENCHMARKS

VATEX is multilingual, large, linguistically complex, and diverse dataset in terms of both video and natural language descriptions. It has two tasks for video-and-language research: (1) Multilingual Video Captioning, aimed at describing a video in various languages with a compact unified captioning model, and (2) Video-guided Machine Translation, to translate a source language description into the target language using the video information as additional spatiotemporal context.

95 PAPERS • 3 BENCHMARKS

The ActivityNet-QA dataset contains 58,000 human-annotated QA pairs on 5,800 videos derived from the popular ActivityNet dataset. The dataset provides a benchmark for testing the performance of VideoQA models on long-term spatio-temporal reasoning.

85 PAPERS • 2 BENCHMARKS

UAVDT is a large scale challenging UAV Detection and Tracking benchmark (i.e., about 80, 000 representative frames from 10 hours raw videos) for 3 important fundamental tasks, i.e., object DETection (DET), Single Object Tracking (SOT) and Multiple Object Tracking (MOT).

80 PAPERS • 1 BENCHMARK

The MovieQA dataset is a dataset for movie question answering. to evaluate automatic story comprehension from both video and text. The data set consists of almost 15,000 multiple choice question answers obtained from over 400 movies and features high semantic diversity. Each question comes with a set of five highly plausible answers; only one of which is correct. The questions can be answered using multiple sources of information: movie clips, plots, subtitles, and for a subset scripts and DVS.

79 PAPERS • 1 BENCHMARK

NExT-QA is a VideoQA benchmark targeting the explanation of video contents. It challenges QA models to reason about the causal and temporal actions and understand the rich object interactions in daily activities. It supports both multi-choice and open-ended QA tasks. The videos are untrimmed and the questions usually invoke local video contents for answers.

74 PAPERS • 3 BENCHMARKS

The How2 dataset contains 13,500 videos, or 300 hours of speech, and is split into 185,187 training, 2022 development (dev), and 2361 test utterances. It has subtitles in English and crowdsourced Portuguese translations.

73 PAPERS • 2 BENCHMARKS

A large-scale multi-object tracking dataset for human tracking in occlusion, frequent crossover, uniform appearance and diverse body gestures. It is proposed to emphasize the importance of motion analysis in multi-object tracking instead of mainly appearance-matching-based diagram.

61 PAPERS • 1 BENCHMARK

We present the MSP-IMPROV corpus, a multimodal emotional database, where the goal is to have control over lexical content and emotion while also promoting naturalness in the recordings. Studies on emotion perception often require stimuli with fixed lexical content, but that convey different emotions. These stimuli can also serve as an instrument to understand how emotion modulates speech at the phoneme level, in a manner that controls for coarticulation. Such audiovisual data are not easily available from natural recordings. A common solution is to record actors reading sentences that portray different emotions, which may not produce natural behaviors. We propose an alternative approach in which we define hypothetical scenarios for each sentence that are carefully designed to elicit a particular emotion. Two actors improvise these emotion-specific situations, leading them to utter contextualized, non-read renditions of sentences that have fixed lexical content and convey different emot

61 PAPERS • 2 BENCHMARKS

Multimodal Opinionlevel Sentiment Intensity (MOSI) contains: (1) multimodal observations including transcribed speech and visual gestures as well as automatic audio and visual features, (2) opinion-level subjectivity segmentation, (3) sentiment intensity annotations with high coder agreement, and (4) alignment between words, visual and acoustic features.

58 PAPERS • 1 BENCHMARK

This YouTube dataset is a sampling from thousands of User Generated Content (UGC) as uploaded to YouTube distributed under the Creative Commons license. This dataset was created in order to assist in the advancement of video compression and quality assessment research of UGC videos.

55 PAPERS • 1 BENCHMARK

EmoryNLP comprises 97 episodes, 897 scenes, and 12,606 utterances, where each utterance is annotated with one of the seven emotions borrowed from the six primary emotions in the Willcox (1982)’s feeling wheel, sad, mad, scared, powerful, peaceful, joyful, and a default emotion of neutral.

52 PAPERS • 1 BENCHMARK

SEED-Bench consists of 19K multiple choice questions with accurate human annotations (~6 larger than existing benchmarks), which spans 12 evaluation dimensions including the comprehension of both the image and video modality.

51 PAPERS • NO BENCHMARKS YET

Room-Across-Room (RxR) is a multilingual dataset for Vision-and-Language Navigation (VLN) for Matterport3D environments. In contrast to related datasets such as Room-to-Room (R2R), RxR is 10x larger, multilingual (English, Hindi and Telugu), with longer and more variable paths, and it includes and fine-grained visual groundings that relate each word to pixels/surfaces in the environment.

42 PAPERS • 1 BENCHMARK

Assembly101 is a new procedural activity dataset featuring 4321 videos of people assembling and disassembling 101 "take-apart" toy vehicles. Participants work without fixed instructions, and the sequences feature rich and natural variations in action ordering, mistakes, and corrections. Assembly101 is the first multi-view action dataset, with simultaneous static (8) and egocentric (4) recordings. Sequences are annotated with more than 100K coarse and 1M fine-grained action segments, and 18M 3D hand poses. We benchmark on three action understanding tasks: recognition, anticipation and temporal segmentation. Additionally, we propose a novel task of detecting mistakes. The unique recording format and rich set of annotations allow us to investigate generalization to new toys, cross-view transfer, long-tailed distributions, and pose vs. appearance. We envision that Assembly101 will serve as a new challenge to investigate various activity understanding problems.

38 PAPERS • 4 BENCHMARKS

XD-Violence is a large-scale audio-visual dataset for violence detection in videos.

36 PAPERS • 1 BENCHMARK

AVSpeech is a large-scale audio-visual dataset comprising speech clips with no interfering background signals. The segments are of varying length, between 3 and 10 seconds long, and in each clip the only visible face in the video and audible sound in the soundtrack belong to a single speaking person. In total, the dataset contains roughly 4700 hours of video segments with approximately 150,000 distinct speakers, spanning a wide variety of people, languages and face poses.

35 PAPERS • NO BENCHMARKS YET

There exist previous works [6, 10] that constructed referring segmentation datasets for videos. Gavrilyuk et al. [6] extended the A2D [33] and J-HMDB [9] datasets with natural sentences; the datasets focus on describing the ‘actors’ and ‘actions’ appearing in videos, therefore the instance annotations are limited to only a few object categories corresponding to the dominant ‘actors’ performing a salient ‘action’. Khoreva et al. [10] built a dataset based on DAVIS [25], but the scales are barely sufficient to learn an end-to-end model from scratch

34 PAPERS • 3 BENCHMARKS

The Query-based Video Highlights (QVHighlights) dataset is a dataset for detecting customized moments and highlights from videos given natural language (NL). It consists of over 10,000 YouTube videos, covering a wide range of topics, from everyday activities and travel in lifestyle vlog videos to social and political activities in news videos. Each video in the dataset is annotated with: (1) a human-written free-form NL query, (2) relevant moments in the video w.r.t. the query, and (3) five-point scale saliency scores for all query-relevant clips.

27 PAPERS • 4 BENCHMARKS

The Wide Multi Channel Presentation Attack (WMCA) database consists of 1941 short video recordings of both bonafide and presentation attacks from 72 different identities. The data is recorded from several channels including color, depth, infra-red, and thermal.

27 PAPERS • 1 BENCHMARK

The Multi-Object and Segmentation (MOTS) benchmark [2] consists of 21 training sequences and 29 test sequences. It is based on the KITTI Tracking Evaluation 2012 and extends the annotations to the Multi-Object and Segmentation (MOTS) task. To this end, we added dense pixel-wise segmentation labels for every object. We evaluate submitted results using the metrics HOTA, CLEAR MOT, and MT/PT/ML. We rank methods by HOTA [1]. Our development kit and GitHub evaluation code provide details about the data format as well as utility functions for reading and writing the label files. (adapted for the segmentation case). Evaluation is performed using the code from the TrackEval repository.

26 PAPERS • 1 BENCHMARK

OpenLane is the first real-world and the largest scaled 3D lane dataset to date. The dataset collects valuable contents from public perception dataset Waymo Open Dataset and provides lane&closest-in-path object(CIPO) annotation for 1000 segments. In short, OpenLane owns 200K frames and over 880K carefully annotated lanes. The OpenLane Dataset is publicly released to aid the research community in making advancements in 3D perception and autonomous driving technology.

26 PAPERS • 2 BENCHMARKS

VALUE is a Video-And-Language Understanding Evaluation benchmark to test models that are generalizable to diverse tasks, domains, and datasets. It is an assemblage of 11 VidL (video-and-language) datasets over 3 popular tasks: (i) text-to-video retrieval; (ii) video question answering; and (iii) video captioning. VALUE benchmark aims to cover a broad range of video genres, video lengths, data volumes, and task difficulty levels. Rather than focusing on single-channel videos with visual information only, VALUE promotes models that leverage information from both video frames and their associated subtitles, as well as models that share knowledge across multiple tasks.

24 PAPERS • NO BENCHMARKS YET

The large-scale MUSIC-AVQA dataset of musical performance contains 45,867 question-answer pairs, distributed in 9,288 videos for over 150 hours. All QA pairs types are divided into 3 modal scenarios, which contain 9 question types and 33 question templates. Finally, as an open-ended problem of our AVQA tasks, all 42 kinds of answers constitute a set for selection.

22 PAPERS • 1 BENCHMARK

UVO is a new benchmark for open-world class-agnostic object segmentation in videos. Besides shifting the problem focus to the open-world setup, UVO is significantly larger, providing approximately 8 times more videos compared with DAVIS, and 7 times more mask (instance) annotations per video compared with YouTube-VOS and YouTube-VIS. UVO is also more challenging as it includes many videos with crowded scenes and complex background motions. Some highlights of the dataset include:

22 PAPERS • 3 BENCHMARKS

An open-ended VideoQA benchmark that aims to: i) provide a well-defined evaluation by including five correct answer annotations per question and ii) avoid questions which can be answered without the video.

22 PAPERS • 2 BENCHMARKS

MAD (Movie Audio Descriptions) is an automatically curated large-scale dataset for the task of natural language grounding in videos or natural language moment retrieval. MAD exploits available audio descriptions of mainstream movies. Such audio descriptions are redacted for visually impaired audiences and are therefore highly descriptive of the visual content being displayed. MAD contains over 384,000 natural language sentences grounded in over 1,200 hours of video, and provides a unique setup for video grounding as the visual stream is truly untrimmed with an average video duration of 110 minutes. 2 orders of magnitude longer than legacy datasets.

21 PAPERS • 2 BENCHMARKS

Action Genome Question Answering (AGQA) is a benchmark for compositional spatio-temporal reasoning. AGQA contains 192M unbalanced question answer pairs for 9.6K videos. It also contains a balanced subset of 3.9M question answer pairs, 3 orders of magnitude larger than existing benchmarks, that minimizes bias by balancing the answer distributions and types of question structures.

20 PAPERS • NO BENCHMARKS YET

Dataset Statistics: The statistics of our dataset are summarized and compared with the largest existing dataset (DR(eye)VE) [1] in Table 1. Our dataset was collected using videos selected from a publicly available, large-scale, crowd-sourced driving video dataset, BDD100k [30, 31]. BDD100K contains human-demonstrated dashboard videos and time-stamped sensor measurements collected during urban driving in various weather and lighting conditions. To efficiently collect attention data for critical driving situations, we specifically selected video clips that both included braking events and took place in busy areas (see supplementary materials for technical details). We then trimmed videos to include 6.5 seconds prior to and 3.5 seconds after each braking event. It turned out that other driving actions, e.g., turning, lane switching and accelerating, were also included. 1,232 videos (=3.5 hours) in total were collected following these procedures. Some example images from our dataset are sh

20 PAPERS • NO BENCHMARKS YET

CholecT50 is a dataset of endoscopic videos of laparoscopic cholecystectomy surgery introduced to enable research on fine-grained action recognition in laparoscopic surgery. It is annotated with triplet information in the form of <instrument, verb, target>. The dataset is a collection of 50 videos consisting of 45 videos from the Cholec80 dataset and 5 videos from an in-house dataset of the same surgical procedure.

20 PAPERS • 7 BENCHMARKS

Toyota Smarthome Trimmed has been designed for the activity classification task of 31 activities. The videos were clipped per activity, resulting in a total of 16,115 short RGB+D video samples. activities were performed in a natural manner. As a result, the dataset poses a unique combination of challenges: high intra-class variation, high-class imbalance, and activities with similar motion and high duration variance. Activities were annotated with both coarse and fine-grained labels. These characteristics differentiate Toyota Smarthome Trimmed from other datasets for activity classification.

19 PAPERS • 1 BENCHMARK

The Ryerson Audio-Visual Database of Emotional Speech and Song (RAVDESS) contains 7,356 files (total size: 24.8 GB). The database contains 24 professional actors (12 female, 12 male), vocalizing two lexically-matched statements in a neutral North American accent. Speech includes calm, happy, sad, angry, fearful, surprise, and disgust expressions, and song contains calm, happy, sad, angry, and fearful emotions. Each expression is produced at two levels of emotional intensity (normal, strong), with an additional neutral expression. All conditions are available in three modality formats: Audio-only (16bit, 48kHz .wav), Audio-Video (720p H.264, AAC 48kHz, .mp4), and Video-only (no sound). Note, there are no song files for Actor_18.

18 PAPERS • 6 BENCHMARKS

Subjective video quality assessment (VQA) strongly depends on semantics, context, and the types of visual distortions. A lot of existing VQA databases cover small numbers of video sequences with artificial distortions. When testing newly developed Quality of Experience (QoE) models and metrics, they are commonly evaluated against subjective data from such databases, that are the result of perception experiments. However, since the aim of these QoE models is to accurately predict natural videos, these artificially distorted video databases are an insufficient basis for learning. Additionally, the small sizes make them only marginally usable for state-of-the-art learning systems, such as deep learning. In order to give a better basis for development and evaluation of objective VQA methods, we have created a larger datasets of natural, real-world video sequences with corresponding subjective mean opinion scores (MOS) gathered through crowdsourcing. ​ We took YFCC100m as a baseline databas

17 PAPERS • 1 BENCHMARK

SUTD-TrafficQA (Singapore University of Technology and Design - Traffic Question Answering) is a dataset which takes the form of video QA based on 10,080 in-the-wild videos and annotated 62,535 QA pairs, for benchmarking the cognitive capability of causal inference and event understanding models in complex traffic scenarios. Specifically, the dataset proposes 6 challenging reasoning tasks corresponding to various traffic scenarios, so as to evaluate the reasoning capability over different kinds of complex yet practical traffic events.

16 PAPERS • 1 BENCHMARK

SeaDronesSee is a large-scale data set aimed at helping develop systems for Search and Rescue (SAR) using Unmanned Aerial Vehicles (UAVs) in maritime scenarios. Building highly complex autonomous UAV systems that aid in SAR missions requires robust computer vision algorithms to detect and track objects or persons of interest. This data set provides three sets of tracks: object detection, single-object tracking and multi-object tracking. Each track consists of its own data set and leaderboard.

16 PAPERS • 3 BENCHMARKS