AI Paper Atlas

Research map / Benchmarks / ActivityNet

ActivityNet leaderboard

31 papers at ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR report results on ActivityNet. Best accuracy: 94.7 (UniFormerV2-L (32 frames), ICCV 2023).

Numbers are as each paper reports them, read by AI from its results tables and not checked by hand. Papers can differ in setting (training data, model size, extra data); the topic column shows what each paper is about. Open the paper before you compare or quote a number.

ActivityNet · accuracy

21 papers, each with its best reported number for its own method; higher is better.

#accuracy ↑MethodPaperTopic
194.7UniFormerV2-L (32 frames)
CLIP-ViT-L
UniFormerV2: Unlocking the Potential of Image ViTs for Video Understanding
ICCV 2023 · Table 13
Video segmentation and tracking
293.3Ours ViT-L (336↑)
CLIP ViT-L
Revisiting Classifier: Transferring Vision-Language Models for Video Recognition
AAAI 2023 · Table: Comparisons with SOTAs
CLIP and zero-shot
392.1SGClip fine-tuned with 5% of training data
SGClip
ESCA: Contextualizing Embodied Agents via Scene-Graph Generation
NeurIPS 2025 · text
Multimodal LLMs
492.05SurdCRL
InternVideo2-B
Introducing Decomposed Causality with Spatiotemporal Object-Centric Representation for Video Classification
AAAI 2026 · Table 1
Video segmentation and tracking
588.2MultiTrain (312p)
MViTv2
Multi-dataset Training of Transformers for Robust Action Recognition
NeurIPS 2022 · Table 1
Action recognition
686.8DG-P3D
ResNet-101 + K400
Optimization Planning for 3D ConvNets
ICML 2021 · Table: ActivityNet comparisons
Action recognition
780.3ImgAud-Skimming
ResNet-152
Listen to Look: Action Recognition by Previewing Audio
CVPR 2020 · ActivityNet SOTA comparison ta
Action recognition
876.67VideoTrace-R1-7B
Qwen2.5-VL-7B
VideoTrace-R1: Long Video-based Retrieval-Augmented Generation via Reinforcement Learning
ICML 2026 · Table 1
Video-language understanding
964.02VHD
3DResNet-50 → VGG16
Hilbert Distillation for Cross-Dimensionality Networks
NeurIPS 2022 · Table 1
3D generation
1062.5LLaVA-Prime-7B
Qwen2-7B
Learning Beyond Still Frames: Scaling Vision-Language Models with Video
ICCV 2025 · Table 8
Video-language understanding
11 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

ActivityNet · mAP

12 papers, each with its best reported number for its own method; higher is better.

#mAP ↑MethodPaperTopic
196.9Ours ViT-L (336↑)
CLIP ViT-L
Revisiting Classifier: Transferring Vision-Language Models for Video Recognition
AAAI 2023 · Table: Comparisons with SOTAs
CLIP and zero-shot
291.97SMS
TimeSFormer
Search-Map-Search: A Frame Selection Paradigm for Action Recognition
CVPR 2023 · Table 5
Action recognition
389.9ImgAud-Skimming
R(2+1)D-152
Listen to Look: Action Recognition by Previewing Audio
CVPR 2020 · ActivityNet SOTA comparison ta
Action recognition
480.7Uni-AdaFocus (128^2) w/o sample-wise dynamic
MN2+RN
Uni-AdaFocus: Spatial-Temporal Dynamic Computation for Video Recognition
TPAMI 2024 · Table 1
Action recognition
579.5AdaFocusV3 (matched to OCSampler GFLOPs)
MobileNet-V2 + ResNet-50
AdaFocusV3: On Unified Spatial-Temporal Dynamic Video Recognition
ECCV 2022 · Table 1
Action recognition
679AdaFocusV2-MN2/RN (128²)
MobileNet-V2 + ResNet
AdaFocus V2: End-to-End Training of Spatial Dynamic Networks for Video Recognition
CVPR 2022 · Table 1
Video segmentation and tracking
778.6ViLaMP
Qwen2-7B
Scaling Video-Language Models to 10K Frames via Hierarchical Differential Distillation
ICML 2025 · Action recognition table
Video-language understanding
875.9Two Branch + Spatial Navigation
ResNet50
Look More but Care Less in Video Recognition
NeurIPS 2022 · Table: Ablation study on navig
Action recognition
974.8VideoIQ
ResNet-50
Dynamic Network Quantization for Efficient Video Inference
ICCV 2021 · Table 1
Action recognition
1074.75TSM+MCA
TSM
Don't Judge by the Look: Towards Motion Coherent Video Representation
ICLR 2024 · Appendix, Table: Evaluation of
Video segmentation and tracking
2 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

Not affiliated with any conference, journal or the benchmark's authors. Corrections: contact@latenticlabs.com.