Research map / Detection and segmentation / CNN features
Relation Distillation Networks for Video Object Detection
Jiajun Deng, Yingwei Pan, Ting Yao, Wengang Zhou, Houqiang Li, Tao Mei
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 |
|---|---|---|---|
| add within-frame object relation module (ablation) | Faster R-CNN · ImageNet VID val | mAP (%): 75.4 -> 78.5 | Improved |
| add cross-frame relation in basic stage (ablation) | Faster R-CNN · ImageNet VID val | mAP (%): 75.4 -> 80.9 | Improved |
| add advanced stage relation distillation (ablation) | RDN basic stage · ImageNet VID val | mAP (%): 80.9 -> 81.8 | Improved |
| replace within-frame relation with cross-frame (spatio-temporal) relation (ablation) | Faster R-CNN + relation · ImageNet VID val | mAP (%): 78.5 -> 80.9 | Improved |
| add Box Linking with Relations post-processing | RDN (ResNet-101) · ImageNet VID val | mAP (%): 81.8 -> 83.8 | Improved |
| add Box Linking with Relations post-processing | RDN (ResNeXt-101-64x4d) · ImageNet VID val | mAP (%): 83.2 -> 84.7 | Improved |
| use BLR instead of Tube Linking (ablation) | RDN (ResNet-101) · ImageNet VID val | mAP (%): 83.4 -> 83.8 | Improved |
| use stronger backbone ResNeXt-101-64x4d instead of ResNet-101 | RDN · ImageNet VID val | mAP (%): 81.8 -> 83.2 | Improved |
| increase temporal spanning range T from 3 to 18 (ablation) | RDN · ImageNet VID val | mAP (%): 80.3 -> 81.8 | Improved |
| increase T beyond 18 to 24 (ablation) | RDN · ImageNet VID val | mAP (%): 81.8 -> 81.7 | No clear effect |
| increase relation modules in basic stage from 0 to 2 (ablation) | RDN basic stage · ImageNet VID val | mAP (%): 75.4 -> 80.9 | Improved |
| stack more than two relation modules (N_b=4) (ablation) | RDN basic stage · ImageNet VID val | mAP (%): 80.9 -> 80.4 | Hurt |
Closest papers by meaning
- Mining Inter-Video Proposal Relations for Video Object Detection (ECCV 2020 · similarity 0.92)
- Relation Networks for Object Detection (CVPR 2018 · similarity 0.90)
- Localize, Assemble, and Predicate: Contextual Object Proposal Embedding for Visual Relation Detection (AAAI 2020 · similarity 0.89)
- Structure Inference Net: Object Detection Using Scene-Level Context and Instance-Level Relationships (CVPR 2018 · similarity 0.89)
- Dense Relation Distillation With Context-Aware Aggregation for Few-Shot Object Detection (CVPR 2021 · similarity 0.88)
- Leveraging Long-Range Temporal Relationships Between Proposals for Video Object Detection (ICCV 2019 · similarity 0.88)
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.
