Research map / Video / Action recognition
PosMLP-Video: Spatial and Temporal Relative Position Encoding for Efficient Video Recognition
Yanbin Hao, Diansong Zhou, Zhicai Wang, Chong‐Wah Ngo, Xiangnan He, Meng Wang
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 |
|---|---|---|---|
| combine PoTGU and PoSGU in parallel instead of PoSGU alone (ablation) | PosMLP-Video-S · Something-Something V1 (no pretraining) | Top-1 accuracy (%): 6.25 -> 46.31 | Improved |
| add temporal gating (PoTGU) in place of spatial-only PoSGU (ablation) | PosMLP-Video-S · Something-Something V1 (no pretraining) | Top-1 accuracy (%): 6.25 -> 25.44 | Improved |
| use joint PoSTGU instead of factorized PoTGU+PoSGU (ablation) | PosMLP-Video-S · Something-Something V1 (no pretraining) | Top-1 / GFLOPs: 40.89 vs 46.31; 103.09 vs 40.49 GFLOPs | Mixed |
| parallel PoTGU+PoSGU instead of cascaded PoTGU->PoSGU (ablation) | PosMLP-Video-S · Something-Something V1 (no pretraining) | Top-1 accuracy (%): 44.11 -> 46.31 | Improved |
| pretrain on ImageNet-1K (ablation) | PosMLP-Video-S · Something-Something V1 | Top-1 accuracy (%): 46.31 -> 52.24 | Improved |
| use positional units instead of gMLP SGU+TGU (ablation) | PosMLP-Video-S · Something-Something V1 (no pretraining) | Top-1 accuracy (%): 42.94 -> 46.31 | Improved |
| use PoSGU instead of temporal convolution+PoSGU (TConv) (ablation) | PosMLP-Video-S · Something-Something V1 (no pretraining) | Top-1 accuracy (%): 42.25 -> 46.31 | Improved |
| use temporal window 16 instead of 8 (ablation) | PosMLP-Video-S · Something-Something V1 (no pretraining) | Top-1 accuracy (%): 40.73 -> 46.31 | Improved |
| enlarge spatial window in stage 1 to 28x28 (ablation) | PosMLP-Video-S · Something-Something V1 (no pretraining) | Top-1 accuracy (%): 46.31 -> 46.29 | No clear effect |
| use group number g=(8,16,32,64) instead of g=(1,1,1,1) (ablation) | PosMLP-Video-S · Something-Something V1 (no pretraining) | Top-1 accuracy (%): 43.90 -> 46.31 | Improved |
| raise group number to (12,24,48,96) (ablation) | PosMLP-Video-S · Something-Something V1 (no pretraining) | Top-1 accuracy (%): 46.31 -> 45.96 | Hurt |
| use learnable relative position encoding (LRPE) instead of GQPE (ablation) | PosMLP-Video-S · Something-Something V1 (no pretraining) | Top-1 accuracy (%): 38.06 -> 46.31 | Improved |
Closest papers by meaning
- VLG: General Video Recognition with Web Textual Knowledge (IJCV 2024 · similarity 0.86)
- Rethinking Spatiotemporal Feature Learning: Speed-Accuracy Trade-offs in Video Classification (ECCV 2018 · similarity 0.86)
- MorphMLP: An Efficient MLP-Like Backbone for Spatial-Temporal Representation Learning (ECCV 2022 · similarity 0.86)
- Perception Encoder: The best visual embeddings are not at the output of the network (NeurIPS 2025 · similarity 0.85)
- Video-FocalNets: Spatio-Temporal Focal Modulation for Video Action Recognition (ICCV 2023 · similarity 0.85)
- MotionBench: Benchmarking and Improving Fine-grained Video Motion Understanding for Vision Language Models (CVPR 2025 · similarity 0.85)
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.
