Research map / Video / Action recognition
E2E-LOAD: End-to-End Long-form Online Action Detection
Shuqiang Cao, Weixin Luo, Bairui Wang, Wei Zhang, Lin Ma
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add long-term compression and long-short-term fusion (ablation) | E2E-LOAD baseline · THUMOS'14 | mAP: 71.2 -> 72.2 | Improved |
| add efficient inference token re-usage (ablation) | E2E-LOAD baseline · THUMOS'14 | FPS: 9.1 -> 19.5 | Improved |
| add efficient inference token re-usage (ablation) | E2E-LOAD baseline · THUMOS'14 | mAP: 71.2 -> 71.5 | Improved |
| use progressive long-term compression x2,x2,x1,x1 instead of x4,x2,x1 (ablation) | LC module · THUMOS'14 | mAP: 70.8 -> 72.4 | Improved |
| fuse long-term into short-term at middle layer 5 instead of early layer 1 (ablation) | LSF · THUMOS'14 | mAP: 70.5 -> 72.4 | Improved |
| fuse at layer 5 instead of late layer 11 (ablation) | LSF · THUMOS'14 | mAP: 71.5 -> 72.4 | Improved |
| cross-attention fusion instead of self-attention fusion (ablation) | LSF · THUMOS'14 | mAP: 70.3 -> 72.4 | Improved |
| use MViT pooling attention instead of Video Swin shifted window (ablation) | Short-term modeling · THUMOS'14 | mAP: 64.7 -> 71.5 | Improved |
| use MViT pooling attention instead of MeMViT (ablation) | Short-term modeling · THUMOS'14 | mAP: 70.9 -> 71.5 | Improved |
| switch LSTR backbone from TSN to MViT features | LSTR RGB-only · THUMOS'14 | mAP: 56.8 -> 60.7 | Improved |
| end-to-end training of LSTR instead of frozen features | LSTR RGB-only · THUMOS'14 | mAP: 56.8 -> 59.2 | Improved |
| add optical flow input to feature-based LSTR with MViT | LSTR · THUMOS'14 | mAP: 60.7 -> 71.2 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- GateHUB — GateHUB: Gated History Unit With Background Suppression for Online Action Detection (CVPR 2022) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- E2E-LOAD++: End-to-end Long-form Online Action Detection and Beyond (TPAMI 2026 · similarity 0.94)
- MiniROAD: Minimal RNN Framework for Online Action Detection (ICCV 2023 · similarity 0.87)
- Context-Enhanced Memory-Refined Transformer for Online Action Detection (CVPR 2025 · similarity 0.86)
- End-to-End Temporal Action Detection with 1B Parameters Across 1000 Frames (CVPR 2024 · similarity 0.85)
- Does Video-Text Pretraining Help Open-Vocabulary Online Action Detection? (NeurIPS 2024 · similarity 0.84)
- Online Video Understanding: OVBench and VideoChat-Online (CVPR 2025 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
