Research map / Video / Action recognition
Coarse-Fine Networks for Temporal Activity Detection in Videos
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use learned Grid Pool and Multi-stage Fusion instead of the SlowFast_det two-stream baseline | X3D two-stream detector · Charades localization | mAP (%): +2.79% mAP | Helped |
| Use Coarse-Fine instead of Fine-Fine by adding Grid Pool (ablation) | X3D-M detector · Charades localization | mAP (%): 24.43 -> 25.10; +0.67% mAP | Helped |
| Use Multi-stage Fusion instead of late fusion only (ablation) | Fine-Fine · Charades localization | mAP (%): 21.84 -> 22.65; +0.81% mAP | Helped |
| Add multiple fine-feature abstraction levels to intermediate fusion (ablation) | Fine-Fine · Charades localization | mAP (%): 22.50 -> 22.65; +0.15% mAP | Helped |
| Include temporal dimension and Gaussian location calibration in fusion (ablation) | Fine-Fine · Charades localization | mAP (%): 19.86 -> 22.65; +2.79% mAP | Helped |
| Add self-attention filtering to fine features before Multi-stage Fusion (ablation) | Fine-Fine · Charades localization | mAP (%): 21.42 -> 22.65; +1.23% mAP | Helped |
| Replace max pooling with learned Grid Pool (ablation) | Coarse-only · Charades localization | mAP (%): 16.21 -> 18.12; +1.91% mAP | Helped |
| Replace average pooling with learned Grid Pool (ablation) | Coarse-only · Charades localization | mAP (%): 16.64 -> 18.12; +1.48% mAP | Helped |
| Use uniform temporal striding instead of max pooling (ablation) | Coarse-only · Charades localization | mAP (%): 16.21 -> 17.49; +1.28% mAP | Helped |
| Use uniform temporal striding instead of average pooling (ablation) | Coarse-only · Charades localization | mAP (%): 16.64 -> 17.49; +0.85% mAP | Helped |
| Increase Grid Pool downsampling from α=1/4 to α=1/8 at T=128 (ablation) | Coarse-only · Charades localization | mAP (%): 18.12 -> 11.88 | Hurt |
| Increase the input length from T=128 to T=256 at α=1/4 (ablation) | Coarse-only · Charades localization | mAP (%): 18.12 -> 18.16; +0.02% mAP | No clear effect |
Closest papers by meaning
- Temporal Convolutional Networks for Action Segmentation and Detection (CVPR 2017 · similarity 0.90)
- Temporal Query Networks for Fine-Grained Video Understanding (CVPR 2021 · similarity 0.89)
- Weakly Supervised Action Localization by Sparse Temporal Pooling Network (CVPR 2018 · similarity 0.88)
- MS-TCT: Multi-Scale Temporal ConvTransformer for Action Detection (CVPR 2022 · similarity 0.88)
- CDC: Convolutional-De-Convolutional Networks for Precise Temporal Action Localization in Untrimmed Videos (CVPR 2017 · similarity 0.88)
- FineParser: A Fine-grained Spatio-temporal Action Parser for Human-centric Action Quality Assessment (CVPR 2024 · similarity 0.87)
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.