Research map / Video / Action recognition
Two-Stream Region Convolutional 3D Network for Temporal Activity Detection
Huijuan Xu, Abir Das, Kate Saenko
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 |
|---|---|---|---|
| use two-way buffer instead of one-way buffer (ablation) | Single-stream R-C3D · THUMOS'14 | mAP@0.5 (%): 27.0 -> 28.9 | Improved |
| add optical flow stream (sum fusion) (ablation) | Single-stream R-C3D · THUMOS'14 | mAP@0.5 (%): 28.9 -> 33.4 | Improved |
| sum instead of concat fusion of the two streams (ablation) | Two-stream R-C3D · THUMOS'14 | mAP@0.5 (%): 33.1 -> 33.4 | No clear effect |
| add optical flow stream (sum fusion) (ablation) | Single-stream R-C3D · THUMOS'14 proposals | Proposal AUC (%): 26.36 -> 29.71 | Improved |
| add online hard example mining (ablation) | Single-stream R-C3D · THUMOS'14 | mAP@0.5 (%): 28.9 -> 35.8 | Improved |
| add online hard example mining (ablation) | Two-stream R-C3D (Sum) · THUMOS'14 | mAP@0.5 (%): 33.4 -> 36.1 | Improved |
| add optical flow stream (sum fusion) (ablation) | Single-stream R-C3D · ActivityNet 1.3 validation | mAP@0.5 (%): 26.8 -> 26.5 | No clear effect |
| add optical flow stream (sum fusion) (ablation) | Single-stream R-C3D · ActivityNet 1.3 proposals | Proposal AUC (%): 56.7 -> 55.6 | No clear effect |
| add online hard example mining (ablation) | Single-stream R-C3D · ActivityNet 1.3 validation | mAP@0.5 (%): 26.8 -> 27.7 | Improved |
| add online hard example mining (ablation) | Single-stream R-C3D · ActivityNet 1.3 validation | Average mAP (%): 12.7 -> 15.4 | Improved |
| train on train+val instead of train only (ablation) | Single-stream R-C3D · ActivityNet 1.3 test | mAP@0.5 (%): 26.8 -> 28.4 | Improved |
| evaluate short activities (under 13 s) instead of long (over 53 s) | Single-stream R-C3D + OHEM · ActivityNet 1.3 validation | mAP@0.5 (%): 39.2 -> 8.3 | Hurt |
Closest papers by meaning
- R-C3D: Region Convolutional 3D Network for Temporal Activity Detection (ICCV 2017 · similarity 0.97)
- Rethinking Spatiotemporal Feature Learning: Speed-Accuracy Trade-offs in Video Classification (ECCV 2018 · similarity 0.90)
- T-C3D: Temporal Convolutional 3D Network for Real-Time Action Recognition (AAAI 2018 · similarity 0.89)
- Temporal Convolutional Networks for Action Segmentation and Detection (CVPR 2017 · similarity 0.89)
- ZSTAD: Zero-Shot Temporal Activity Detection (CVPR 2020 · similarity 0.89)
- HARDVS: Revisiting Human Activity Recognition with Dynamic Vision Sensors (AAAI 2024 · 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.
