Research map / Video / Video-language understanding
Dense video captioning
Localizes and describes multiple events using joint modeling, memory retrieval and weak supervision.
65 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Video-language understanding, keyword cluster “grounding · captioning · sentence”
State History Benchmarks Tried Baselines Checklist Preprints Who Export
Where it stands
Still moving The best reported result is still improving on 1 of 6 tracked benchmarks.
13 papers in 2023–24 and 19 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 5.8% to 2.6%.
Open questions the data points to
- YouCook2 (CIDEr): best went from 59.35 (up to 2024) to 71.84 in 2025 — HiCM²: Hierarchical Compact Memory Modeling for Dense Video Captioning (AAAI 2025 · 6 citations)
What looks settled
- YouCook2 (METEOR): best up to 2024 9.41, best in 2025–26 6.11 from 3 papers, at most 0.2% better, if at all.
- YouCook2 (SODA c): best up to 2024 7.97, best in 2025–26 5.39 from 2 papers, at most 0.2% better, if at all.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2016 · Reading the Videos: Temporal Labeling for Crowdsourced Time-Sync Videos Based on Semantic Embedding (AAAI 2016 · 54 citations)
- 2017 · Lip Reading Sentences in the Wild (CVPR 2017 · 613 citations)
- 2018 · End-to-End Dense Video Captioning With Masked Transformer (CVPR 2018 · 578 citations)
- 2019 · To Find Where You Talk: Temporal Sentence Localization in Video with Attention Based Location Regression (AAAI 2019 · 363 citations)
- 2020 · Learning 2D Temporal Adjacent Networks for Moment Localization with Natural Language (AAAI 2020 · 498 citations)
- 2021 · Boundary Proposal Network for Two-stage Natural Language Video Localization (AAAI 2021 · 162 citations)
- 2022 · Weakly Supervised Video Moment Localization with Contrastive Negative Sample Mining (AAAI 2022 · 104 citations)
- 2023 · Vid2Seq: Large-Scale Pretraining of a Visual Language Model for Dense Video Captioning (CVPR 2023 · 245 citations)
- 2024 · VTimeLLM: Empower LLM to Grasp Video Moments (CVPR 2024 · 119 citations)
- 2025 · Dense Audio-Visual Event Localization Under Cross-Modal Consistency and Multi-Temporal Granularity Collaboration (AAAI 2025 · 13 citations)
- 2026 · TimeChat-Captioner: Scripting Multi-Scene Videos with Time-Aware and Structural Audio-Visual Captions (ICML 2026 · 0 citations)
Earliest papers here
- Reading the Videos: Temporal Labeling for Crowdsourced Time-Sync Videos Based on Semantic Embedding (AAAI 2016 · 54 citations)
- Video Semantic Clustering with Sparse and Incomplete Tags (AAAI 2016 · 19 citations)
- Lip Reading Sentences in the Wild (CVPR 2017 · 613 citations)
Benchmark records
The best number each year that beat every earlier one, among papers of this approach reporting their own method. Numbers are extracted by AI from each paper's text; check the paper before citing.
YouCook2 · CIDEr (higher is better)
6 papers, 2024–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2024 | 59.35 | DIBS: Enhancing Dense Video Captioning with Unlabeled Videos via Pseudo Boundary Enrichment and Online Refinement (CVPR 2024 · 14 citations) |
| 2025 | 71.84 | HiCM²: Hierarchical Compact Memory Modeling for Dense Video Captioning (AAAI 2025 · 6 citations) |
Charades-STA · R@1, IoU=0.5 (higher is better)
5 papers, 2020–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 23.58 | Weakly-Supervised Video Moment Retrieval via Semantic Completion Network (AAAI 2020 · 172 citations) |
| 2021 | 64.06 | Structured Multi-Level Interaction Network for Video Moment Localization via Language Query (CVPR 2021 · 100 citations) |
Charades-STA · R@1, IoU=0.7 (higher is better)
5 papers, 2020–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 9.97 | Weakly-Supervised Video Moment Retrieval via Semantic Completion Network (AAAI 2020 · 172 citations) |
| 2021 | 40.75 | Structured Multi-Level Interaction Network for Video Moment Localization via Language Query (CVPR 2021 · 100 citations) |
YouCook2 · METEOR (higher is better)
5 papers, 2024–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2024 | 9.41 | DIBS: Enhancing Dense Video Captioning with Unlabeled Videos via Pseudo Boundary Enrichment and Online Refinement (CVPR 2024 · 14 citations) |
ActivityNet Captions · mIoU (higher is better)
4 papers, 2020–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 39.8 | Rethinking the Bottom-Up Framework for Query-Based Video Localization (AAAI 2020 · 172 citations) |
| 2021 | 42.11 | Boundary Proposal Network for Two-stage Natural Language Video Localization (AAAI 2021 · 162 citations) |
YouCook2 · SODA c (higher is better)
4 papers, 2024–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2024 | 7.97 | DIBS: Enhancing Dense Video Captioning with Unlabeled Videos via Pseudo Boundary Enrichment and Online Refinement (CVPR 2024 · 14 citations) |
What has been tried
55 papers here report 617 measured changes (a component added, removed or swapped, compared with a baseline): helped 76%, no clear effect 5%, hurt 17%, mixed 2%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Use image rather than video data for Stage 1 feature alignment | ActivityNet temporal grounding · R@0.3 | 33.5 -> 36.1 | Helped | VTimeLLM: Empower LLM to Grasp Video Moments (CVPR 2024 · 119 citations) |
| Replace clustering memory with temporal pooling | ActivityNet, T=64 · CIDEr | 30.6 -> 26.8 | Hurt | Streaming Dense Video Captioning (CVPR 2024 · 48 citations) |
| add cross-modal memory retrieval | ActivityNet Captions · CIDEr | 31.24 -> 33.01 | Helped | Do You Remember? Dense Video Captioning with Cross-Modal Memory Retrieval (CVPR 2024 · 32 citations) |
| Add generated pseudo boundaries | YouCook2 · CIDEr | 9.07 -> 21.93 | Helped | DIBS: Enhancing Dense Video Captioning with Unlabeled Videos via Pseudo Boundary Enrichment and Online Refinement (CVPR 2024 · 14 citations) |
| Add both C2F and F2C blocks | UnAV-100 · mAP (Avg.) | 47.9 -> 49.2 | Helped | Dense Audio-Visual Event Localization Under Cross-Modal Consistency and Multi-Temporal Granularity Collaboration (AAAI 2025 · 13 citations) |
| Add unified relative position tokens to Seq2Time | YouCook2 · F1 | 15.7 -> 16.2 | Helped | Seq2Time: Sequential Knowledge Transfer for Video LLM Temporal Grounding (CVPR 2025 · 8 citations) |
| add hierarchical compact memory retrieval with LLM summarization | YouCook2 · CIDEr | 66.29 -> 71.84 | Helped | HiCM²: Hierarchical Compact Memory Modeling for Dense Video Captioning (AAAI 2025 · 6 citations) |
| Add PEI to initialize queries with pseudo-event locations | YouCook2 · CIDEr | 31.15 -> 34.26 | Helped | Event-Equalized Dense Video Captioning (CVPR 2025 · 6 citations) |
| use 250 rather than 300 seed concepts | Charades-STA · R@0.3 | 49.21 -> 50.98 | Helped | Commonsense for Zero-Shot Natural Language Video Localization (AAAI 2024 · 5 citations) |
| Add VSOR-CoT during fine-tuning | MVS · CC | 0.700 -> 0.714 | Helped | CaRDiff: Video Salient Object Ranking Chain of Thought Reasoning for Saliency Prediction with Diffusion (AAAI 2025 · 4 citations) |
| add video-to-text retrieval | ActivityNet Captions, learned proposals · CIDEr | 27.74 -> 31.32 | Helped | Exploring Temporal Event Cues for Dense Video Captioning in Cyclic Co-Learning (AAAI 2025 · 4 citations) |
| Add video-narration contrastive training (VNA) | VISOR-NVOS test · J &F | 28.1 -> 36.0 | Helped | Learning to Segment Referred Objects from Narrated Egocentric Videos (CVPR 2024 · 4 citations) |
Baseline numbers that disagree
The same baseline on the same benchmark and metric, reported more than 5% apart by different papers here. Training budgets, data splits and reimplementations differ, so state which number you compare with.
| Baseline | Benchmark | Metric | Reported | Papers |
|---|---|---|---|---|
| DRN | Charades-STA | R@5, IoU=0.7 | 55.38 – 60.05 | 2 papers, e.g. AAAI 2023 and CVPR 2021 |
What reviewers will expect
Datasets and baselines named in the 12 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- YouCook2 · 6 of 12 papers
- ActivityNet Captions · 5 of 12 papers
- Charades-STA · 2 of 12 papers
- ViTT · 2 of 12 papers
Baselines
- PDVC · 5 of 12 papers
- Vid2Seq · 4 of 12 papers
- VTimeLLM · 3 of 12 papers
- DCE · 2 of 12 papers
- DVC · 2 of 12 papers
- DIBS · 2 of 12 papers
- CM^2 · 2 of 12 papers
- TimeChat · 2 of 12 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 32 papers; from OpenAlex and the papers' affiliations).
- Tsinghua University · 2 papers
- Peking University · 2 papers
- Peng Cheng Laboratory · 2 papers
- Xi’an Jiaotong University · 2 papers
- Shandong University · 2 papers
- University of Science and Technology of China · 2 papers
- Hefei University of Technology · 2 papers
- Electronics and Telecommunications Research Institute · 2 papers
- Kyung Hee University · 2 papers
- The University of Hong Kong · 2 papers
Use it
Download BibTeX of 39 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Temporal video grounding (88 papers)
- Encoder-decoder video captioning (52 papers)
Elsewhere in the topic
- Long video understanding (236 papers)
- Video LLMs with fine-grained understanding (229 papers)
- Event and anomaly reasoning in video (184 papers)
- Video-language pretraining (82 papers)
- Video question answering (64 papers)
- Text-video retrieval (59 papers)
- Egocentric video understanding (46 papers)
- Video moment retrieval (41 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).