Research map / Detection and segmentation / CNN features
Patchwork: A Patch-Wise Attention Network for Efficient Object Detection and Segmentation in Video Streams
Yuning Chai
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 Q-learned attention instead of random policy (ablation) | Patchwork M=4 N=2 · ImageNet VID val (policy ablation, 1/4 cost) | mAP@0.5: 52.5 -> 54.3 | Improved |
| add Q-learned attention instead of scanning policy (ablation) | Patchwork M=2 N=1 vs M=4 N=2 · ImageNet VID val (policy ablation, 1/4 cost) | mAP@0.5: 51.7 -> 54.3 | Improved |
| add Q-learned attention instead of random policy (ablation) | Patchwork M=4 N=1 · ImageNet VID val (policy ablation, 1/16 cost) | mAP@0.5: 37.0 -> 41.6 | Improved |
| add Q-learned attention instead of scanning policy (ablation) | Patchwork M=4 N=1 · ImageNet VID val (policy ablation, 1/16 cost) | mAP@0.5: 37.9 -> 41.6 | Improved |
| add Patchwork Cell to scanning policy (ablation) | Patchwork · ImageNet VID val (Patchwork Cell ablation) | mAP@0.5: 39.5 -> 51.7 | Improved |
| add Patchwork Cell to DQN policy (ablation) | Patchwork · ImageNet VID val (Patchwork Cell ablation) | mAP@0.5: 43.2 -> 54.3 | Improved |
| use Patchwork cells with random attention instead of single-frame model | Patchwork M=4 N=2 · ImageNet VID val (policy ablation, 1/4 cost) | mAP@0.5: 49.3 -> 52.5 | Improved |
| use Patchwork M=4 N=2 at roughly 25% computation instead of full single-frame model | MobileNetV2 SSD · ImageNet VID val | mAP@0.5: 54.7 -> 54.3 | No clear effect |
| reinvest saved compute in depth multiplier 1.4 and flip ensembling | Patchwork M=4 N=2 · ImageNet VID val | mAP@0.5: 54.7 -> 58.7 | Improved |
| use Patchwork M=4 N=2 at about 1/4 computation | MobileNetV2 segmenter · DAVIS 2016 unsupervised | J mean: 63.6 -> 62.1 | Hurt |
| use Patchwork M=4 N=2 at about 1/4 computation | MobileNetV2 segmenter · DAVIS 2016 unsupervised | F mean: 59.5 -> 58.8 | Hurt |
| use Patchwork M=4 N=1 instead of single-frame interval=16 delay=15 at similar cost | MobileNetV2 segmenter · DAVIS 2016 unsupervised | J mean: 27.5 -> 42.2 | Improved |
Closest papers by meaning
- Learning Unsupervised Video Object Segmentation Through Visual Attention (CVPR 2019 · similarity 0.87)
- Learning Video Object Segmentation With Visual Memory (ICCV 2017 · similarity 0.87)
- Object Guided External Memory Network for Video Object Detection (ICCV 2019 · similarity 0.87)
- LiVOS: Light Video Object Segmentation with Gated Linear Matching (CVPR 2025 · similarity 0.87)
- Region-Based Convolutional Networks for Accurate Object Detection and Segmentation (TPAMI 2016 · similarity 0.87)
- Temporal Perceiver: A General Architecture for Arbitrary Boundary Detection (TPAMI 2023 · similarity 0.86)
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.
