Research map / Video / Action recognition
Deep Analysis of CNN-Based Spatio-Temporal Representations for Action 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 |
|---|---|---|---|
| remove temporal pooling from I3D (ablation) | I3D-ResNet50 · Kinetics-400 | top-1 accuracy: +1.12 | Helped |
| remove temporal pooling from I3D (ablation) | I3D-ResNet50 · Something-Something V2 | top-1 accuracy: +6.02 | Helped |
| add a ResNet50 backbone instead of InceptionV1 (ablation) | I3D · Kinetics-400 | top-1 accuracy: +4.09 | Helped |
| add a ResNet50 backbone instead of InceptionV1 (ablation) | I3D · Something-Something V2 | top-1 accuracy: -0.13 | No clear effect |
| add squeeze-excitation modules (ablation) | I3D-ResNet50 · Kinetics-400 | top-1 accuracy: +1.18 | Helped |
| use uniform instead of dense frame sampling (ablation) | I3D-ResNet50 · Kinetics-400 | top-1 accuracy: 76.61 -> 76.07 | Hurt |
| use dense instead of uniform frame sampling (ablation) | TAM-ResNet50 · Kinetics-400 | top-1 accuracy: 76.45 -> 76.18 | Hurt |
| use uniform instead of dense frame sampling (ablation) | SlowFast-ResNet50-8×8 · Kinetics-400 | top-1 accuracy: 76.40 -> 71.85 | Hurt |
| add temporal pooling (ablation) | I3D, S3D, and TAM models · Mini-Kinetics | top-1 accuracy | Hurt |
| add temporal pooling (ablation) | TSN · Mini-SSV2 | top-1 accuracy: >20% | Helped |
| add temporal pooling (ablation) | TSN-ResNet50 · Kinetics-400 | top-1 accuracy: +5.1% | Helped |
| keep TAM modules only in the top half of the network rather than all layers (ablation) | TAM-ResNet18 · Mini-SSV2 | top-1 accuracy: 59.1 -> 59.7 | Helped |
Closest papers by meaning
- T-C3D: Temporal Convolutional 3D Network for Real-Time Action Recognition (AAAI 2018 · similarity 0.93)
- A Closer Look at Spatiotemporal Convolutions for Action Recognition (CVPR 2018 · similarity 0.93)
- ACTION-Net: Multipath Excitation for Action Recognition (CVPR 2021 · similarity 0.91)
- Can Spatiotemporal 3D CNNs Retrace the History of 2D CNNs and ImageNet? (CVPR 2018 · similarity 0.91)
- MiCT: Mixed 3D/2D Convolutional Tube for Human Action Recognition (CVPR 2018 · similarity 0.91)
- Spatiotemporal Fusion in 3D CNNs: A Probabilistic View (CVPR 2020 · similarity 0.90)
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.