Research map / Video / Action recognition
Watch Only Once: An End-to-End Video Action Detection Framework
Shoufa Chen, Peize Sun, Enze Xie, Chongjian Ge, Jiannan Wu, Lan Ma, Jiajun Shen, Ping Luo
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 |
|---|---|---|---|
| replace off-the-shelf person detector with end-to-end head (no embeddings) (ablation) | SlowOnly two-stage baseline · AVA val | mAP@0.5: 20.3 -> 16.9 | Hurt |
| replace off-the-shelf person detector with end-to-end head (no embeddings) (ablation) | SlowOnly two-stage baseline · AVA val | GFLOPs: 136.8 -> 65.1 | Improved |
| add spatial embedding (ablation) | naive E2E · AVA val | mAP@0.5: 16.9 -> 20.4 | Improved |
| add temporal embedding (ablation) | naive E2E · AVA val | mAP@0.5: 16.9 -> 18.5 | Improved |
| add both spatial and temporal embeddings (ablation) | naive E2E · AVA val | mAP@0.5: 16.9 -> 21.3 | Improved |
| add both embeddings, versus two-stage baseline | WOO vs SlowOnly two-stage · AVA val | mAP@0.5: 20.3 -> 21.3 | Improved |
| use cross-attention fusion instead of summation (ablation) | WOO · AVA val | mAP@0.5: 20.5 -> 21.3 | Improved |
| use cross-attention fusion instead of concatenation (ablation) | WOO · AVA val | mAP@0.5: 20.5 -> 21.3 | Improved |
| use FPN instead of dilated res5 (ablation) | WOO · AVA val | mAP@0.5: 20.8 -> 21.3 | Improved |
| use FPN instead of dilated res5 (ablation) | WOO · AVA val | actor detection AP: 74.4 -> 75.6 | Improved |
| plug FPN with key-frame input into two-stage baseline (ablation) | SlowFast · AVA val | mAP@0.5: 20.3 -> 19.1 | Hurt |
| WOO versus two-stage SlowFast R50 | SlowFast R50 · AVA v2.1 val | mAP@0.5: 24.7 -> 25.2 | Improved |
Closest papers by meaning
- Classification Matters: Improving Video Action Detection with Class-Specific Attention (ECCV 2024 · similarity 0.91)
- ActionSwitch: Class-agnostic Detection of Simultaneous Actions in Streaming Videos (ECCV 2024 · similarity 0.91)
- End-To-End Learning of Action Detection From Frame Glimpses in Videos (CVPR 2016 · similarity 0.90)
- Graph Convolutional Module for Temporal Action Localization in Videos (TPAMI 2021 · similarity 0.90)
- Learning To Refactor Action and Co-Occurrence Features for Temporal Action Localization (CVPR 2022 · similarity 0.90)
- PointTAD: Multi-Label Temporal Action Detection with Learnable Query Points (NeurIPS 2022 · similarity 0.90)
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.
