Research map / Video / Action recognition
Learning Self-Similarity in Space and Time As Generalized Motion for Video Action Recognition
Heeseung Kwon, Manjin Kim, Suha Kwak, Minsu Cho
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 a SELFY block | TSM-R50 (16 frames) · Something-Something V1 | top-1 accuracy (%): 47.3 -> 54.3 | Improved |
| add a SELFY block | TSM-R50 (16 frames) · Something-Something V2 | top-1 accuracy (%): 61.2 -> 65.7 | Improved |
| add a SELFY block | TSM-R50 (16 frames) · Diving-48 | top-1 accuracy (%): 38.8 -> 41.6 | Improved |
| add a SELFY block | TSM-R50 (8 frames) · FineGym Gym288 | top-1 accuracy (%): 47.9 -> 49.5 | Improved |
| add a SELFY block | TSM (16 frames) · Kinetics-400 | top-1 accuracy (%): 74.7 -> 77.1 | Improved |
| lengthen temporal offset range from {1} to {-3..3} (ablation) | SELFYNet on TSM-R18 · Something-Something V1 (TSM-R18, 8 frames) | top-1 accuracy (%): 47.1 -> 48.6 | Improved |
| add temporal range {-3..3} instead of {1} (ablation) | SELFYNet on TSN-R18 · Something-Something V1 (TSM-R18, 8 frames) | top-1 accuracy (%): 39.7 -> 46.9 | Improved |
| use convolution instead of soft-argmax for feature extraction (ablation) | SELFYNet on TSM-R18 · Something-Something V1 (TSM-R18, 8 frames) | top-1 accuracy (%): 44.0 -> 46.7 | Improved |
| use convolution instead of MLP for feature integration (ablation) | SELFYNet on TSM-R18 · Something-Something V1 (TSM-R18, 8 frames) | top-1 accuracy (%): 47.2 -> 48.4 | Improved |
| place SELFY block after res3 rather than res5 (ablation) | SELFYNet on TSM-R18 · Something-Something V1 (TSM-R18, 8 frames) | top-1 accuracy (%): 42.8 -> 48.4 | Improved |
| add STSS features Z to visual features V (ablation) | SELFYNet on TSM-R18 · Something-Something V1 (TSM-R18, 8 frames) | top-1 accuracy (%): 45.5 -> 48.4 | Improved |
| use SELFY instead of self-attention (MHSA) (ablation) | TSM-R18 · Something-Something V1 (TSM-R18, 8 frames) | top-1 accuracy (%): 44.0 -> 48.4 | Improved |
Closest papers by meaning
- Self-Supervised Spatio-Temporal Representation Learning for Videos by Predicting Motion and Appearance Statistics (CVPR 2019 · similarity 0.89)
- Searching for Better Spatio-temporal Alignment in Few-Shot Action Recognition (NeurIPS 2022 · similarity 0.88)
- Learning Motion in Feature Space: Locally-Consistent Deformable Convolution Networks for Fine-Grained Action Detection (ICCV 2019 · similarity 0.88)
- Time Is MattEr: Temporal Self-supervision for Video Transformers (ICML 2022 · similarity 0.88)
- Learning SpatioTemporal and Motion Features in a Unified 2D Network for Action Recognition (TPAMI 2022 · similarity 0.88)
- Collaborative Spatiotemporal Feature Learning for Video Action Recognition (CVPR 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.
