Research map / Video / Action recognition
PA3D: Pose-Action 3D Machine for Video Recognition
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 |
|---|---|---|---|
| Use temporal pose convolution instead of TSN-style frame score averaging (ablation) | PA3D joint heatmaps · JHMDB and HMDB (split one) | accuracy: 54.5 -> 59.3; 42.3 -> 46.7 | Helped |
| Use temporal pose convolution instead of 3D convolution (ablation) | PA3D joint heatmaps · JHMDB and HMDB (split one) | accuracy: 55.6 -> 59.3; 45.1 -> 46.7 | Mixed |
| Use dilated temporal pose convolution with global convolution (ablation) | PA3D joint modality · JHMDB and HMDB | accuracy: 59.3 -> 60.1; 46.7 -> 47.8 | Helped |
| Use local temporal pose convolution rather than global temporal pose convolution (ablation) | PA3D joint modality · JHMDB and HMDB | accuracy: 59.3 -> 57.5; 46.7 -> 44.9 | Hurt |
| Set the number of sampled frames to 12 rather than 8 (ablation) | PA3D joint modality · JHMDB and HMDB | accuracy: 55.6 -> 50.4; 46.8 -> 43.2 | Hurt |
| Increase output channels from N=4 to N=8 (ablation) | PA3D joint modality · JHMDB and HMDB | accuracy: 59.6 -> 60.1; 46.3 -> 46.0 | Mixed |
| Use score fusion instead of representation concatenation for joints and parts (ablation) | PA3D pose modalities · JHMDB and HMDB | accuracy: 58.7 -> 65.6; 47.5 -> 50.7 | Mixed |
| Add joint, part, and feature differences to the fused pose modalities (ablation) | PA3D · JHMDB and HMDB | accuracy: 67.5 -> 69.5; 54.1 -> 54.7 | Mixed |
| Fuse PA3D with I3D | I3D · HMDB | accuracy: 80.7 -> 82.1 | Helped |
| Fuse PA3D with RPAN | RPAN · JHMDB | accuracy: 83.9 -> 86.1 | Helped |
| Fuse PA3D with GCN, I3D, and NL I3D | GCN + I3D + NL I3D · Charades | mAP: 40.7 -> 41.0 | Helped |
| Pretrain action CNN on Charades rather than train from scratch (ablation) | PA3D action CNN on HMDB · HMDB | accuracy: around 1-2% | Hurt |
Closest papers by meaning
- T-C3D: Temporal Convolutional 3D Network for Real-Time Action Recognition (AAAI 2018 · similarity 0.90)
- 3D Action Recognition From Novel Viewpoints (CVPR 2016 · similarity 0.89)
- Deep Analysis of CNN-Based Spatio-Temporal Representations for Action Recognition (CVPR 2021 · similarity 0.89)
- 3DV: 3D Dynamic Voxel for Action Recognition in Depth Video (CVPR 2020 · similarity 0.89)
- 2D/3D Pose Estimation and Action Recognition Using Multitask Deep Learning (CVPR 2018 · similarity 0.89)
- 3DInAction: Understanding Human Actions in 3D Point Clouds (CVPR 2024 · similarity 0.86)
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.