Real-world videos are usually long, untrimmed, and contain several actions (events). Traditionally, video understanding has focused on short-term analysis, such as action recognition, object detection/segmentation, or scene understanding in individual video frames or short video clips. However, with the advent of more advanced technologies and the increasing availability of large-scale video datasets, researchers have started exploring long-term video understanding. It involves developing methods that can recognize and understand complex activities, events, or interactions that unfold over longer durations of time. This includes tasks like temporal action detection, dense video captioning, video grounding, future video prediction, and video summarization, among others.
This repository curates a collection of research works specifically focused on long-term videos. This is an active repository and your contributions are always welcome!
- Awesome Long-Term Video Understanding
- HierVL: Learning Hierarchical Video-Language Embeddings, CVPR 2023.
- Procedure-Aware Pretraining for Instructional Video Understanding, CVPR 2023.
- Learning Grounded Vision-Language Representation for Versatile Understanding in Untrimmed Videos, arXiv 2023.
- Temporal Alignment Networks for Long-Term Video, CVPR 2022.
- Long-form video-language pre-training with multimodal temporal contrastive learning, NIPS 2022.
- Learning from Untrimmed Videos: Self-Supervised Video Representation Learning with Hierarchical Consistency, CVPR 2022.
- Unsupervised Pre-training for Temporal Action Localization Tasks, CVPR 2022.
- Hierarchical Self-supervised Representation Learning for Movie Understanding, CVPR 2022.
- Egocentric Video-Language Pretraining, NIPS 2022.
- LocVTP: Video-Text Pre-training for Temporal Localization, ECCV 2022.
- Boundary-sensitive Pre-training for Temporal Localization in Videos, ICCV 2021.
- TSP: Temporally-Sensitive Pretraining of Video Encoders for Localization Tasks, ICCVW 2021.
- COOT: Cooperative Hierarchical Transformer for Video-Text Representation Learning, NIPS 2020.
- MeMViT: Memory-Augmented Multiscale Vision Transformer for Efficient Long-Term Video Recognition, CVPR 2022.
- Long-term feature banks for detailed video understanding, CVPR 2019.
- Multi-agent reinforcement learning based frame sampling for effective untrimmed video recognition, ICCV 2019.
- TimeChat: A Time-sensitive Multimodal Large Language Model for Long Video Understanding, CVPR 2024.
- OmniVid: A Generative Framework for Universal Video Understanding, CVPR 2024.
- Language Repository for Long Video Understanding, 2024.
- VTimeLLM: Empower LLM to Grasp Video Moments, CVPR 2024.
- Momentor: Advancing Video Large Language Model with Fine-Grained Temporal Reasoning, Arxiv 2024.
- MA-LMM: Memory-Augmented Large Multimodal Model for Long-Term Video Understanding, CVPR 2024.
- VideoMosaic: Connecting the Temporal Dots in Long Videos for LLMs, CVPR 2024.
- MovieChat: From Dense Token to Sparse Memory for Long Video Understanding, CVPR 2024.
- TimeChat: A Time-sensitive Multimodal Large Language Model for Long Video Understanding, CVPR 2024.
- MVBench: A Comprehensive Multi-modal Video Understanding Benchmark (VideoChat2), CVPR 2024.
- LLMs Meet Long Video: Advancing Long Video Comprehension with An Interactive Visual Adapter in LLMs, 2024.
- LSTP: Language-guided Spatial-Temporal Prompt Learning for Long-form Video-Text Understanding, 2024.
- [Agent] MM-Narrator: Narrating Long-form Videos with Multimodal In-Context Learning, CVPR 2024.
- [Agent] VideoAgent: Long-form Video Understanding with Large Language Model as Agent, 2024.
- [Agent] VideoAgent: A Memory-augmented Multimodal Agent for Video Understanding, 2024.
- (TODO)
Resources:
- Github repo: Awesome Weakly Supervised Temporal Action Localization
- Github repo: Awesome Temporal Action Localization
- Github repo: Awesome Temporal Action Detection Temporal Action Proposal Generation
- [survey] Deep Learning-Based Action Detection in Untrimmed Videos: A Survey, TPAMI 2023.
- [survey] A survey on temporal action localization, IEEE Access 2020.
Representitive Papers (before 2023):
- ActionFormer: Localizing Moments of Actions with Transformers, ECCV 2022.
- G-tad: Sub-graph localization for temporal action detection, CVPR 2020.
- Relaxed Transformer Decoders for Direct Action Proposal Generation, ICCV 2021.
- Fast learning of temporal action proposal via dense boundary generator, AAAI 2020.
- BMN: boundary-matching network for temporal action proposal generation, ICCV 2019.
- Graph convolutional networks for temporal action localization, ICCV 2019.
- Rethinking the faster r-cnn architecture for temporal action localization, CVPR 2018.
- BSN: boundary sensitive network for temporal action proposal generation, ECCV 2018.
- SST: Single-Stream Temporal Action Proposals, CVPR 2017.
- Temporal action detection with structured segment networks, ICCV 2017.
- R-c3d: Region convolutional 3d network for temporal activity detection, ICCV 2017.
- CDC: Convolutional-De-Convolutional Networks for Precise Temporal Action Localization in Untrimmed Videos, CVPR 2017.
- TURN TAP: Temporal Unit Regression Network for Temporal Action Proposals, ICCV 2017.
- Single Shot Temporal Action Detection, ACM MM 2017.
- Temporal action localization in untrimmed videos via multi-stage CNNs, CVPR 2016.
- Temporal Localization of Actions with Actoms, TPAMI 2014.
Latest Papers (after 2023)
- ETAD: Training Action Detection End to End on a Laptop, CVPR 2023.
- TriDet: Temporal Action Detection With Relative Boundary Modeling, CVPR 2023.
- Distilling Vision-Language Pre-training to Collaborate with Weakly-Supervised Temporal Action Localization, CVPR 2023.
- Re2TAL: Rewiring Pretrained Video Backbones for Reversible Temporal Action Localization, CVPR 2023.
- Boosting Weakly-Supervised Temporal Action Localization With Text Information, CVPR 2023.
- Proposal-based Multiple Instance Learning for Weakly-supervised Temporal Action Localization, CVPR 2023.
- Deepsegmenter: Temporal action localization for detecting anomalies in untrimmed naturalistic driving videos, CVPR 2023.
- Dense-Localizing Audio-Visual Events in Untrimmed Videos: A Large-Scale Benchmark and Baseline, CVPR 2023. [code]
- Towards Long Form Audio-visual Video Understanding, Arxiv 2023.
- Audio-Visual Event Localization in Unconstrained Videos, ECCV 2018.
- Video ReCap: Recursive Captioning of Hour-Long Videos, CVPR 2024.
- Learning Grounded Vision-Language Representation for Versatile Understanding in Untrimmed Videos, arXiv 2023.
- Vid2Seq: Large-Scale Pretraining of a Visual Language Model for Dense Video Captioning, CVPR 2023. [code]
- Unifying event detection and captioning as sequence generation via pre-training, ECCV 2022.
- End-to-end Dense Video Captioning as Sequence Generation, Coling 2022.
- End-to-End Dense Video Captioning with Parallel Decoding, ICCV 2021. [code]
- Sketch, Ground, and Refine: Top-Down Dense Video Captioning, CVPR 2021.
- Streamlined Dense Video Captioning, CVPR 2019.
- Bidirectional Attentive Fusion with Context Gating for Dense Video Captioning, CVPR 2018.
- End-to-End Dense Video Captioning with Masked Transformer, CVPR 2018.
- Weakly Supervised Dense Event Captioning in Videos, NIPS 2018.
- Dense-Captioning Events in Videos, ICCV 2017.
Video Paragraph Captioning:
- Adversarial Inference for Multi-Sentence Video Description, CVPR 2019
- COOT: Cooperative Hierarchical Transformer for Video-Text Representation Learning, NIPS 2020.
- Multimodal Pretraining for Dense Video Captioning, AACL 2020.
- MART: Memory-Augmented Recurrent Transformer for Coherent Video Paragraph Captioning, ACL 2020.
- Towards Diverse Paragraph Captioning for Untrimmed Videos, CVPR 2021.
- Video Paragraph Captioning as a Text Summarization Task, ACL 2021.
Resources:
- [survey] Temporal Sentence Grounding in Videos: A Survey and Future Directions, TPAMI 2022.
- [survey] A survey on temporal sentence grounding in videos, ACM TOMM 2023.
- [survey] A survey on video moment localization, ACM Computing Surveys, 2023.
- Github repo: Awesome Video Grounding
Representative Papers (before 2023):
- Negative sample matters: A renaissance of metric learning for temporal grounding, AAAI 2022.
- Dense events grounding in video, AAAI 2021.
- Local-global video-text interactions for temporal grounding, CVPR 2020.
- Dense regression network for video grounding, CVPR 2020.
- Learning 2d temporal adjacent networks for moment localization with natural language, AAAI 2020.
- To Find Where You Talk: Temporal Sentence Localization in Video with Attention Based Location Regression, AAAI 2020.
- Man: Moment alignment network for natural language moment retrieval via iterative graph adjustment, CVPR 2019.
- Semantic conditioned dynamic modulation for temporal sentence grounding in videos, NIPS 2019.
- Temporally grounding natural sentence in video, EMNLP 2018.
- TALL: Temporal activity localization via language query, ICCV 2017.
Latest Papers (after 2023):
- Learning Grounded Vision-Language Representation for Versatile Understanding in Untrimmed Videos, Arxiv 2023.
- ProTeG´e: Untrimmed Pretraining for Video Temporal Grounding by Video ´ Temporal Grounding, CVPR 2023.
- Text-visual prompting for efficient 2d temporal video grounding, CVPR 2023.
- DeCo : Decomposition and Reconstruction for Compositional Temporal Grounding via Coarse-to-Fine Contrastive Ranking, CVPR 2023.
- Hierarchical Semantic Correspondence Networks for Video Paragraph Grounding, CVPR 2023.
- Iterative Proposal Refinement for Weakly-Supervised Video Grounding, CVPR 2023.
- Collaborative Static and Dynamic Vision-Language Streams for Spatio-Temporal Video Grounding, CVPR 2023.
- You Can Ground Earlier than See: An Effective and Efficient Pipeline for Temporal Sentence Grounding in Compressed Videos, CVPR 2023.
- Text-Visual Prompting for Efficient 2D Temporal Video Grounding, CVPR 2023.
- Weakly Supervised Temporal Sentence Grounding with Uncertainty-Guided Self-training, CVPR 2023.
- Towards Generalisable Video Moment Retrieval: Visual-Dynamic Injection to Image-Text Pre-Training, CVPR 2023.
- Hierarchical Video-Moment Retrieval and Step-Captioning, CVPR 2023.
- [survey] A Review on Deep Learning Techniques for Video Prediction, TPAMI, 2020. [paper]
- Revisiting Hierarchical Approach for Persistent Long-Term Video Prediction, ICLR 2021.
- Hierarchical Long-term Video Prediction Without Supervision, ICML 2018. [paper]
- Learning to Generate Long-term Future via Hierarchical Prediction, ICML 2017.
- [Action quality assessment] LOGO: A Long-Form Video Dataset for Group Action Quality Assessment, CVPR 2023.
- [Spatiotemporal Grounding] Relational Space-Time Query in Long-Form Videos, CVPR 2023.
- [Tracking] XMem: Long-Term Video Object Segmentation with An Atkinson-Shiffrin Memory Model, ECCV 2022.
- [VQA] Long-Term Video Question Answering Via Multimodal Hierarchical Memory Attentive Networks, TCSVT 2021.
- [Video summarization] Video Summarization with Long Short-term Memory, ECCV 2016.
Latest Papers (after 2023) [ Dataset ] Panda-70M: Captioning 70M Videos with Multiple Cross-Modality Teachers, CVPR 2024. [ Dataset ] MovieLLM: Enhancing Long Video Understanding with AI-Generated Movies, 2024. A novel framework designed to create synthetic, high-quality data for long videos. This framework leverages the power of GPT-4 and text-to-image models to generate detailed scripts and corresponding visuals.
| Dataset | Annotation | Source | Number | Duration | Tasks | link | Date Released |
|---|---|---|---|---|---|---|---|
| ActivityNet 1.3 | timestamps + action | Youtube | 20k | - | Action Localization | ||
| ActivityNet Captions | timestamps + captions | Youtube | 20k | - | Dense captioning, video grounding | ||
| THUMOS | timestamps + action | - | - | - | Action Localization | ||
| YouCook2 | timestamps + captions | Cooking Videos | - | - | Dense captioning | ||
| MovieNet | timestamps + captions + place/action/style tags | Movies | 1.1k | >2h | movie understanding | MovieNet | 2020 |
| Charades | timestamps + action labels | Daily Activities | 9.8k | 30s | action recognition, action localization | Charades | 2017 |
| Charades-STA | timestamps + captions | Daily Activities | 9.8k | 30s | video grounding | Charades-STA | 2017 |
| TACoS | timestamps + captions | Cooking | - | - | video grounding | [TACoS] | |
| DiDeMo | |||||||
| TACoS Multi-Level | timestamps + captions | Cooking | - | - | Dense captioning | TACoS Multi-Level | |
| VIOLIN | YouTube and TV shows | 6.7k / 15.8k clips | Video-and-Language Inference | VIOLIN | |||
| HowTo100M | boundaries + captions | 1.22M/136M clips | 582h | pretraining | - | ||
| YT-temporal180M | boundaries + captions | - | - | - | pretraining | - | |
| EPIC-KITCHENS | - | - | - | - | - | epic-kitchens | - |
| HD_VILA_100M | |||||||
| FineGym | - | - | - | - | - | FineGym | - |
| MAD | - | - | - | - | - | MAD | - |
| MPII-MD | - | - | - | - | - | MPII-MD | - |
| M-VAD | - | - | - | - | - | M-VAD | - |
| HACS | - | - | - | - | - | HACS | - |
| PKU-MMD | - | - | - | - | - | PKU-MMD | - |
| MultiTHUMOS | - | - | - | - | - | MultiTHUMOS | - |
| VideoLT | - | - | - | - | - | VideoLT | 2021 |
| Ego4D | - | - | - | - | - | Ego4D | 2021 |
| ActivityNet Entities | - | - | - | - | - | Anet-Entity | 2020 |
| MovieQA | - | - | - | - | - | MoiveQA(expired) | 2018 |
| MultiSports | |||||||
| FineAction | - | - | - | - | - | FineAction | 2022 |
| MUSES | - | - | - | - | - | MUSES | 2021 |
| XD-Violence | - | - | - | - | - | XD-Violence | 2020 |
| TVR | - | - | - | - | - | TVR | 2020 |
| HIREST | - | - | - | - | - | HIREST | 2023 |
| QVHighlights | - | - | - | - | - | QVHighlights | 2020 |
| ANetQA | - | - | - | - | - | ANetQA | 2023 |
- https://github.com/v-iashin/video_features
- https://github.com/zjr2000/Untrimmed-Video-Feature-Extractor
- https://github.com/hobincar/pytorch-video-feature-extractor
Latest Papers (after 2023)
- TempCompass: Do Video LLMs Really Understand Videos? 2024.
- MVBench: A Comprehensive Multi-modal Video Understanding Benchmark, CVPR 2024.
- How Good is my Video LMM? Complex Video Reasoning and Robustness Evaluation Suite for Video-LMMs, 2024
- add more details of datasets
- add a plot to illustrate differences between untrimmed videos and trimmed videos, and also between different localization tasks
