Skip to the content.

Video Annotation Tasks

This category contains annotation task designs for video understanding research, including action recognition, temporal localization, video summarization, and scene understanding.

Run these designs in Potato. See the Potato documentation to configure temporal and frame-level video annotation.

Subcategories

Action Recognition

Design Description Reference
activitynet-temporal-localization ActivityNet is a large-scale video benchmark for human activity understanding, with untrimmed YouTube… Caba Heilbron et al., CVPR 2015
ava-atomic-actions AVA spatio-temporally localizes 80 atomic actions on people in movie clips, labeled at 1 Hz keyframes Gu et al., CVPR 2018
breakfast-actions Fine-grained temporal action segmentation of breakfast preparation activities Kuehne et al., CVPR 2014
charades-activity-segmentation Multi-label temporal activity segmentation in indoor home videos Sigurdsson et al., ECCV 2016
epic-kitchens-egocentric Annotate fine-grained actions in egocentric kitchen videos with verb-noun pairs Damen et al., ECCV 2018
finediving-procedure-aqa Fine-grained diving action quality assessment Xu et al., CVPR 2022
finegym-action-segments FineGym is a hierarchical video dataset (CVPR 2020) for fine-grained gymnastics action recognition:… Shao et al., CVPR 2020
fisv-figure-skating-scoring Automatic figure skating scoring as video regression Xu et al., IEEE TCSVT 2020
soccernet-action-spotting SoccerNet-v2 defines 17 action spotting classes (goal, corner, card, foul, substitution, and more) with… Giancola et al., CVPR 2018
thumos14-action-localization Temporal action localization in untrimmed sports videos Jiang et al., ECCV 2014 Workshop

AMI Meeting Multi-Tier Annotation

Design Description Reference
ami-meeting-annotation Multi-tier ELAN-style annotation of multi-party meeting recordings Carletta et al., MLMI 2005

Boundary Detection

Design Description Reference
moviescenes-detection Detect and annotate scene boundaries in movies Rao et al., CVPR 2020
scene-boundary-detection Identify scene boundaries in documentary and narrative videos Sidiropoulos et al., IEEE TCSVT 2011
shot-boundary-detection Detect shot boundaries and classify transition types in broadcast video Smeaton et al., 2010

CHILDES Child Language Multi-Tier Annotation

Design Description Reference
childes-child-language Multi-tier ELAN-style annotation of child-adult interaction videos for language acquisition research MacWhinney, 2000 (Lawrence Erlbaum, 3rd ed.)

CMU-MOSEI: Multimodal Sentiment and Emotion Dataset

Design Description Reference
cmu-mosei-multimodal-sentiment CMU-MOSEI is the largest multimodal dataset for sentiment and emotion analysis, with 23,453 annotated… Zadeh et al., ACL 2018

DGS Corpus Sign Language Multi-Tier Annotation

Design Description Reference
dgs-corpus-sign-language Multi-tier ELAN-style annotation of German Sign Language (DGS) corpus videos Hanke et al., LREC 2020

Ego4D: Egocentric Video Episodic Memory Annotation

Design Description Reference
ego4d-episodic-memory Annotate egocentric (first-person) video for episodic memory tasks including activity segmentation, hand… Grauman et al., CVPR 2022

FineSports: Fine-grained Basketball Action Recognition

Design Description Reference
finesports-action-recognition Fine-grained NBA basketball action recognition Xu et al., CVPR 2024

Harmony4D Human Interaction Tracking

Design Description Reference
harmony4d-interaction-tracking Close-range human interaction tracking and annotation Khirodkar et al., NeurIPS 2024

How2Sign Sign Language Multi-Tier Annotation

Design Description Reference
how2sign-sign-language Multi-tier ELAN-style annotation of continuous American Sign Language videos Duarte et al., CVPR 2021

IEMOCAP: Dyadic Emotion Recognition Dataset

Design Description Reference
iemocap-dyadic-emotion IEMOCAP is a 12-hour audiovisual corpus of two-actor emotional conversations with categorical and… Busso et al., Language Resources and Evaluation 2008

Instructional

Design Description Reference
howto100m-instructional Annotate instructional video clips with step descriptions and visual grounding Miech et al., ICCV 2019
youcook2-instructional YouCook2 contains 2,000 cooking videos across 89 recipes (176 hours), each segmented into recipe steps… Zhou et al., AAAI 2018

MSAD Multi-Scenario Anomaly Detection

Design Description Reference
msad-anomaly-detection Video anomaly detection across multiple scenarios Zhu et al., NeurIPS 2024

SaGA Gesture-Speech Alignment Multi-Tier Annotation

Design Description Reference
saga-gesture-speech Multi-tier ELAN-style annotation of co-speech gestures and their alignment with spoken language Lücking et al., Journal on Multimodal User Interfaces 2013

Scene Understanding

Design Description Reference
movienet-scene-classification Scene classification labeling on MovieNet, a 1,100-movie dataset with scene, place, action, and cinematic-style annotations Huang et al., ECCV 2020

Summarization

Design Description Reference
lsmdc-keyframe-selection Select representative keyframes from movie clips for video description tasks Rohrbach et al., IJCV 2017
summe-summarization Create video summaries by selecting key segments that best represent the content Gygli et al., ECCV 2014
tvsum-summarization Frame-level importance scoring for video summarization Song et al., CVPR 2015
youtube-highlights Detect highlight-worthy moments in domain-specific videos Sun et al., ECCV 2014

Temporal Grounding

Design Description Reference
activitynet-captions ActivityNet Captions pairs 20k untrimmed videos with 100k temporally localized sentence descriptions for… Krishna et al., ICCV 2017
charades-sta-grounding Ground natural language descriptions to video segments Gao et al., ICCV 2017
didemo-moment-retrieval Localizing natural language descriptions to specific video moments Hendricks et al., ICCV 2017
qvhighlights-moment-saliency Query-based moment localization and highlight rating based on the QVHighlights dataset (Lei, Berg, Bansal,… Lei et al., NeurIPS 2021

VBench: Video Generation Quality Benchmark

Design Description Reference
vbench-generation-quality VBench (CVPR 2024) breaks text-to-video quality into 16 evaluation dimensions, from temporal flickering to… Huang et al., CVPR 2024

Video-ChatGPT Video QA Display and Evaluation

Design Description Reference
video-chatgpt-qa-display Video question answering evaluation based on the Video-ChatGPT benchmark Maaz et al., ACL 2024

Video QA

Design Description Reference
mvbench-video-understanding Comprehensive video understanding benchmark with multiple-choice questions, video segment annotation, and… Li et al., CVPR 2024
nextqa-temporal NExT-QA is a video QA benchmark for causal and temporal action reasoning over 5,440 videos and ~52K questions Xiao et al., CVPR 2021

VSTAR Video-grounded Dialogue

Design Description Reference
vstar-video-dialogue Video-grounded dialogue annotation Wang et al., ACL 2023

Quick Start

# Navigate to a specific task
cd video/action-recognition/ava-atomic-actions

# Run with Potato
potato start config.yaml

Annotation Modes

The video_annotation scheme takes a mode setting. In segment mode annotators mark start and end times; in keyframe mode they pick representative frames; in tracking mode they follow an object across frames; and combined mode puts several of these on the same video.

Task Count

Total: 40 video annotation tasks across 21 subcategories