Research map / Video / Action recognition
VideoCapsuleNet: A Simplified Network for Action Detection
Read the paperCheck 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 VideoCapsuleNet instead of the strongest reported baseline | UCF-Sports · UCF-Sports | v-mAP (IoU 0.2): 96.0 -> 97.1 | Helped |
| Use VideoCapsuleNet instead of the strongest reported baseline | J-HMDB · J-HMDB | v-mAP (IoU 0.2): 79.7 -> 95.1 | Helped |
| Use VideoCapsuleNet instead of the strongest reported baseline | UCF-101 (24 classes) · UCF-101 (24 classes) | v-mAP (IoU 0.5): 59.9 -> 80.3 | Helped |
| Add localization loss to classification training (ablation) | VideoCapsuleNet · UCF-101 | accuracy: 62.0 -> 79.0 | Helped |
| Add classification loss to localization training (ablation) | VideoCapsuleNet · UCF-101 | v-mAP (IoU 0.5): 48.1 -> 80.3 | Helped |
| Add reconstruction loss while omitting localization loss (ablation) | VideoCapsuleNet · UCF-101 | accuracy: 62.0 -> 72.2 | Helped |
| Add reconstruction loss alongside classification and localization losses (ablation) | VideoCapsuleNet · UCF-101 | accuracy: 79.0 -> 73.6 | Hurt |
| Add skip connections from 3D Conv1, Conv2, and Conv4 (ablation) | VideoCapsuleNet · UCF-101 | v-mAP (IoU 0.5): 80.3 -> 80.7 | No clear effect |
| Add coordinate addition to the final convolutional capsule votes (ablation) | VideoCapsuleNet · UCF-101 | accuracy: 71.7 -> 79.0 | Helped |
| Add coordinate addition to the final convolutional capsule votes (ablation) | VideoCapsuleNet · UCF-101 | f-mAP (IoU 0.5): 72.9 -> 78.6 | Helped |
| Use 6 rather than 4 convolution layers (ablation) | VideoCapsuleNet · UCF-101 | v-mAP (IoU 0.5): 73.5 -> 80.3 | Helped |
| Use 6 rather than 8 convolution layers (ablation) | VideoCapsuleNet · UCF-101 | accuracy: 71.4 -> 79.0 | Helped |
Closest papers by meaning
- G-TAD: Sub-Graph Localization for Temporal Action Detection (CVPR 2020 · similarity 0.88)
- CDC: Convolutional-De-Convolutional Networks for Precise Temporal Action Localization in Untrimmed Videos (CVPR 2017 · similarity 0.88)
- A Structured Model for Action Detection (CVPR 2019 · similarity 0.87)
- MiCT: Mixed 3D/2D Convolutional Tube for Human Action Recognition (CVPR 2018 · similarity 0.87)
- Temporal Action Localization in Untrimmed Videos via Multi-Stage CNNs (CVPR 2016 · similarity 0.87)
- Temporal Convolutional Networks for Action Segmentation and Detection (CVPR 2017 · 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.