Research map / Video / Action recognition
Towards Efficient Coarse-to-Fine Networks for Action and Gesture Recognition
Niamul Quader, Juwei Lu, Peng Dai, Wei Li
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 |
|---|---|---|---|
| extend TSM to C2F ensemble (classroom learning) | TSM-ResNet50 · Something-Something V1 | top-1 accuracy (%): 47.0 -> 48.4 | Improved |
| extend TSM to C2F ensemble (classroom learning) | TSM-ResNet50 · Something-Something V2 | top-1 accuracy (%): 61.5 -> 62.4 | Improved |
| extend TSM to C2F ensemble | TSM-ResNet50 · Jester | top-1 accuracy (%): 96.3 -> 96.5 | No clear effect |
| add non-local blocks to TSM | TSM-ResNet50 · Something-Something V1 | top-1 accuracy (%): 47.0 -> 41.3 | Hurt |
| add non-local blocks to C2FC,En | C2FC,En · Kinetics-400 | top-1 accuracy (%): 70.5 -> 71.4 | Improved |
| use coarse-exit instead of full ensemble | C2FC · Something-Something V2 | FLOPs and top-1: 62.4 -> 61.8; FLOPs 1.08x -> 0.7x | Mixed |
| feed coarse pathways non-downsampled input instead of C2F decomposition (remove decomposition) (ablation) | C2FC · Something-Something V1 | delta top-1: 1.1 -> 1.4 | Improved |
| add reparameterized FC layer instead of regular FC (ablation) | C2FC · Jester | top-1 accuracy (%): about 0.1 gain | Improved |
| add reparameterized FC layer instead of regular FC (ablation) | C2FC · Kinetics-400 | top-1 accuracy (%): about 0.6 gain | Improved |
| add reparameterized FC layer instead of regular FC (ablation) | C2FC · Something-Something V1 and V2 | top-1 accuracy (%): about 0.3 gain | Improved |
| use additive softmax ensemble of pathways instead of C2F fusion (ablation) | C2F · Jester | top-1 accuracy (%): 96.3 -> 95.3 | Hurt |
| end-to-end multi-stream loss without C2F fusion design (ablation) | C2F · Jester | top-1 accuracy (%): 96.3 -> 96.2 | No clear effect |
Closest papers by meaning
- Sympathy for the Details: Dense Trajectories and Hybrid Classification Architectures for Action Recognition (ECCV 2016 · similarity 0.87)
- Coarse-Fine Networks for Temporal Activity Detection in Videos (CVPR 2021 · similarity 0.87)
- Learning Motion in Feature Space: Locally-Consistent Deformable Convolution Networks for Fine-Grained Action Detection (ICCV 2019 · similarity 0.87)
- Action Recognition With Coarse-to-Fine Deep Feature Integration and Asynchronous Fusion (AAAI 2018 · similarity 0.87)
- Action Recognition With Spatial-Temporal Discriminative Filter Banks (ICCV 2019 · similarity 0.87)
- EAN: Event Adaptive Network for Enhanced Action Recognition (IJCV 2022 · 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.
