Research map / Video / Video segmentation and tracking
Fully Motion-Aware Network for Video Object Detection
Shiyao Wang, Yucong Zhou, Junjie Yan, Zhidong Deng
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 pixel-level calibration to multi-frame aggregation (ablation) | multi-frame aggregation · ImageNet VID val | mAP (%): 73.4 -> 76.5 | Improved |
| add instance-level calibration to multi-frame aggregation (ablation) | multi-frame aggregation · ImageNet VID val | mAP (%): 73.4 -> 77.1 | Improved |
| aggregate multi-frame features by plain averaging without motion guidance (ablation) | single-frame R-FCN · ImageNet VID val | mAP (%): 73.6 -> 73.4 | Hurt |
| add motion pattern reasoning combining pixel and instance calibration (ablation) | instance-level calibration · ImageNet VID val | mAP (%): 77.1 -> 78.1 | Improved |
| combine pixel and instance calibration (ablation) | pixel-level calibration · ImageNet VID val | mAP (%): 76.5 -> 78.1 | Improved |
| plain averaging of frames on fast-moving objects (ablation) | single-frame R-FCN · ImageNet VID val (fast) | mAP (%): 52.2 -> 45.2 | Hurt |
| use instance-level instead of pixel-level calibration on occluded objects (ablation) | MANet · ImageNet VID val (occlusion subset) | mAP (%): 73.0 -> 74.1 | Improved |
| use instance-level instead of pixel-level calibration on non-rigid motion (ablation) | MANet · ImageNet VID val (non-rigid subset) | mAP (%): 52.8 -> 51.6 | Hurt |
| use instance-level instead of pixel-level calibration on rigid motion (ablation) | MANet · ImageNet VID val (rigid subset) | mAP (%): 81.0 -> 81.9 | Improved |
| combine both calibrations instead of best single one on non-rigid motion (ablation) | MANet · ImageNet VID val (non-rigid subset) | mAP (%): 52.8 -> 53.2 | Improved |
| aggregate 4 nearby frames instead of single frame at inference (ablation) | MANet · ImageNet VID val | mAP (%): 73.57 -> 77.58 | Improved |
| increase aggregated frames k from 4 to 12 (ablation) | MANet · ImageNet VID val | mAP (%): 77.58 -> 78.09 | Improved |
Closest papers by meaning
- MoNet: Deep Motion Exploitation for Video Object Segmentation (CVPR 2018 · similarity 0.83)
- Detect to Track and Track to Detect (ICCV 2017 · similarity 0.83)
- Object Detection in Video with Spatiotemporal Sampling Networks (ECCV 2018 · similarity 0.83)
- Objects Do Not Disappear: Video Object Detection by Single-Frame Object Location Anticipation (ICCV 2023 · similarity 0.83)
- Flow-Guided Feature Aggregation for Video Object Detection (ICCV 2017 · similarity 0.82)
- Temporal Context Enhanced Feature Aggregation for Video Object Detection (AAAI 2020 · similarity 0.82)
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.
