Research map / Video / Action recognition
Scene Flow to Action Map: A New Representation for RGB-D Based Action Recognition With Convolutional Neural Networks
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 |
|---|---|---|---|
| Replace hand-designed channel mapping with learned Channel Transform Kernels (ablation) | SFAM rank pooling · ChaLearn LAP IsoGD validation | accuracy: 23.62% -> 27.48% | Helped |
| Use sum accumulation rather than absolute-difference accumulation (ablation) | SFAM · ChaLearn LAP IsoGD validation | accuracy: 18.86% -> 25.83% | Helped |
| Use sum accumulation rather than rank pooling (ablation) | SFAM · ChaLearn LAP IsoGD validation | accuracy: 23.62% -> 25.83% | Helped |
| Fine-tune ImageNet-pretrained weights rather than train from scratch (ablation) | SFAM-D · ChaLearn LAP IsoGD validation | accuracy: 9.23% -> 18.86% | Helped |
| Fine-tune ImageNet-pretrained weights rather than train from scratch (ablation) | SFAM-S · ChaLearn LAP IsoGD validation | accuracy: 18.10% -> 25.83% | Helped |
| Fuse SFAM variant scores by multiplication rather than averaging (ablation) | All SFAM variants · ChaLearn LAP IsoGD validation | accuracy: 34.86% -> 36.27% | Helped |
| Fuse SFAM variant scores by multiplication rather than maximum (ablation) | All SFAM variants · ChaLearn LAP IsoGD validation | accuracy: 33.24% -> 36.27% | Helped |
| Use the sum-accumulation SFAM instead of rank pooling (ablation) | SFAM · M²I single task, side view | accuracy: 79.9% -> 70.1% | Hurt |
| Use the sum-accumulation SFAM instead of rank pooling (ablation) | SFAM · M²I single task, front view | accuracy: 81.8% -> 75.0% | Hurt |
| Fuse SFAM variant scores by multiplication rather than averaging (ablation) | All SFAM variants · M²I single task, side view | accuracy: 88.2% -> 89.4% | Helped |
| Fuse SFAM variant scores by multiplication rather than averaging (ablation) | All SFAM variants · M²I single task, front view | accuracy: 89.1% -> 91.2% | Helped |
| Fuse SFAM variant scores by multiplication rather than averaging (ablation) | All SFAM variants · M²I cross-view, SV → FV | accuracy: 85.3% -> 87.6% | Helped |
Closest papers by meaning
- Optical Flow Guided Feature: A Fast and Robust Motion Representation for Video Action Recognition (CVPR 2018 · similarity 0.90)
- MARS: Motion-Augmented RGB Stream for Action Recognition (CVPR 2019 · similarity 0.90)
- Deep Analysis of CNN-Based Spatio-Temporal Representations for Action Recognition (CVPR 2021 · similarity 0.89)
- Spatiotemporal Residual Networks for Video Action Recognition (NeurIPS 2016 · similarity 0.88)
- Representation Flow for Action Recognition (CVPR 2019 · similarity 0.88)
- T-C3D: Temporal Convolutional 3D Network for Real-Time Action Recognition (AAAI 2018 · 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.