Research map / Detection and segmentation / CNN features
Mining Inter-Video Proposal Relations for Video Object Detection
Mingfei Han, Yali Wang, Xiaojun Chang, Yu Qiao
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 intra-video proposal relation module (ablation) | Faster-RCNN · ImageNet VID val (ResNet-101) | mAP (%): 73.2 -> 80.6 | Improved |
| add inter-video proposal relation module (ablation) | HVR-Net (intra only) · ImageNet VID val (ResNet-101) | mAP (%): 80.6 -> 83.2 | Improved |
| use multi-level triplet selection instead of random video and all-proposal selection (ablation) | HVR-Net · ImageNet VID val (ResNet-101) | mAP (%): 81.0 -> 83.2 | Improved |
| add relation regularization loss (ablation) | HVR-Net · ImageNet VID val (ResNet-101) | mAP (%): 80.0 -> 83.2 | Improved |
| use 3 intra modules instead of 2 (ablation) | HVR-Net · ImageNet VID val (ResNet-101) | mAP (%): 81.8 -> 83.2 | Improved |
| use 2 inter modules instead of 1 (ablation) | HVR-Net · ImageNet VID val (ResNet-101) | mAP (%): 83.2 -> 82.1 | Hurt |
| increase testing frames from 5 to 31 (ablation) | HVR-Net · ImageNet VID val (ResNet-101) | mAP (%): 80.5 -> 83.2 | Improved |
| increase testing frames from 21 to 31 (ablation) | HVR-Net · ImageNet VID val (ResNet-101) | mAP (%): 82.9 -> 83.2 | No clear effect |
| remove inter-video proposal relation module at test time (ablation) | HVR-Net · ImageNet VID val (ResNet-101) | mAP (%) | No clear effect |
| add Seq-NMS post-processing | HVR-Net · ImageNet VID val (ResNet-101) | mAP (%): 83.2 -> 83.8 | Improved |
| add Seq-NMS post-processing | HVR-Net · ImageNet VID val (ResNeXt-101-32x4d) | mAP (%): 84.8 -> 85.5 | Improved |
| use HVR-Net instead of LongRange on fast objects | HVR-Net · ImageNet VID val (Fast objects) | mAP (%): 64.2 -> 66.6 | Improved |
Closest papers by meaning
- Relation Distillation Networks for Video Object Detection (ICCV 2019 · similarity 0.92)
- Leveraging Long-Range Temporal Relationships Between Proposals for Video Object Detection (ICCV 2019 · similarity 0.89)
- Localize, Assemble, and Predicate: Contextual Object Proposal Embedding for Visual Relation Detection (AAAI 2020 · similarity 0.88)
- Beyond Short-Term Snippet: Video Relation Detection With Spatio-Temporal Global Context (CVPR 2020 · similarity 0.87)
- Relation Networks for Object Detection (CVPR 2018 · similarity 0.87)
- Weakly Supervised Object Detection Using Proposal- and Semantic-Level Relationships (TPAMI 2020 · 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.
