Research map / Video / Action recognition
Motion Feature Network: Fixed Motion Filter for Action Recognition
Myunggi Lee, Seungeui Lee, Sungjoon Son, Gyutae Park, Nojun Kwak
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 concatenation motion blocks to TSN (K=7) (ablation) | TSN ResNet-50 · Jester val | top-1 accuracy: 81.0% -> 96.1% | Improved |
| add concatenation motion blocks to TSN (K=7) (ablation) | TSN ResNet-50 · Something-Something val | top-1 accuracy: 8.1% -> 37.3% | Improved |
| add element-wise sum motion blocks to TSN (K=7) (ablation) | TSN ResNet-50 · Something-Something val | top-1 accuracy: 8.1% -> 37.1% | Improved |
| increase training segments K from 3 to 7 (ablation) | MFNet-C50 · Jester val | top-1 accuracy: 90.4% -> 96.1% | Improved |
| increase training segments K from 3 to 7 (ablation) | MFNet-C50 · Something-Something val | top-1 accuracy: 17.4% -> 37.3% | Improved |
| increase training segments K from 3 to 7 (ablation) | TSN baseline · Jester val | top-1 accuracy: 82.4% -> 81.0% | No clear effect |
| use element-wise sum instead of concatenation (ablation) | MFNet-50 · Jester val | top-1 accuracy: +0.2% to +0.6% | Mixed |
| use concatenation instead of element-wise sum (ablation) | MFNet-50 · Something-Something val | top-1 accuracy: +0.2% to +2.8% | Improved |
| increase backbone depth from ResNet-18 to ResNet-101 (ablation) | MFNet-C · Jester val | top-1 accuracy: 96.3% -> 96.7% | No clear effect |
| increase backbone depth from ResNet-18 to ResNet-101 (ablation) | MFNet-C · Something-Something val | top-1 accuracy: 39.4% -> 43.9% | Improved |
| increase backbone depth from ResNet-101 to ResNet-152 (ablation) | MFNet-C · Something-Something val | top-1 accuracy: 43.9% -> 43.0% | No clear effect |
| use random frame interval sampling instead of fixed-time sampling (ablation) | MFNet · Jester and Something-Something | top-1 accuracy | Improved |
Closest papers by meaning
- Learning SpatioTemporal and Motion Features in a Unified 2D Network for Action Recognition (TPAMI 2022 · similarity 0.90)
- Optical Flow Guided Feature: A Fast and Robust Motion Representation for Video Action Recognition (CVPR 2018 · similarity 0.90)
- STM: SpatioTemporal and Motion Encoding for Action Recognition (ICCV 2019 · similarity 0.89)
- Video Modeling With Correlation Networks (CVPR 2020 · similarity 0.89)
- ACTION-Net: Multipath Excitation for Action Recognition (CVPR 2021 · similarity 0.88)
- DynamoNet: Dynamic Action and Motion Network (ICCV 2019 · similarity 0.88)
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.
