Research map / Detection and segmentation / CNN features
Feature Intertwiner for Object Detection
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 feature intertwiner to the detector | Mask R-CNN with ResNet-50-FPN · COCO minival | AP: 32.8 -> 35.2 | Helped |
| add feature intertwiner to the detector | Mask R-CNN with ResNet-50-FPN · COCO minival | AP_S: 12.7 -> 15.3 | Helped |
| use the default proposal assignment rather than assigning by RoI size (ablation) | Intertwiner with ResNet-50-FPN · COCO minival | AP: 33.7 -> 35.2 | Helped |
| assign more proposals to higher levels instead of the default assignment (ablation) | Intertwiner with ResNet-50-FPN · COCO minival | AP: 35.2 -> 32.3 | Hurt |
| use l1 instead of the default l2 intertwiner loss (ablation) | Intertwiner with ResNet-50-FPN · COCO minival | AP: 35.2 -> 34.2 | Hurt |
| use KL divergence instead of the default l2 intertwiner loss (ablation) | Intertwiner with ResNet-50-FPN · COCO minival | AP: 35.2 -> 34.6 | Hurt |
| use a KL divergence loss factor of 10 (ablation) | Intertwiner with ResNet-50-FPN · COCO minival | AP: 35.2 -> 35.6 | Helped |
| use an all-history buffer with equal weights rather than a decay-weight buffer (ablation) | Intertwiner with ResNet-101-FPN · COCO minival | AP: 39.2 -> 40.5 | Helped |
| use a longer partial buffer rather than a buffer of size 2000 (ablation) | Intertwiner with ResNet-101-FPN · COCO minival | AP: 37.3 -> 38.8 | Helped |
| use a unified buffer instead of separate buffers on each level (ablation) | Intertwiner with ResNet-101-FPN · COCO minival | AP: 40.58 vs 40.54 | No clear effect |
| detach the class buffer from gradient updates (ablation) | Intertwiner with ResNet-101-FPN · COCO minival | AP: 40.1 -> 40.5 | Helped |
| linearly combine critic features with the original detection features (ablation) | Intertwiner with ResNet-50-FPN · COCO minival | AP: 34.0 -> 35.2 | Helped |
Closest papers by meaning
- Expanding Low-Density Latent Regions for Open-Set Object Detection (CVPR 2022 · similarity 0.88)
- Distilling Object Detectors via Decoupled Features (CVPR 2021 · similarity 0.87)
- Proper Reuse of Image Classification Features Improves Object Detection (CVPR 2022 · similarity 0.87)
- Distilling Image Classifiers in Object Detectors (NeurIPS 2021 · similarity 0.86)
- DR Loss: Improving Object Detection by Distributional Ranking (CVPR 2020 · similarity 0.86)
- Supervision Interpolation via LossMix: Generalizing Mixup for Object Detection and Beyond (AAAI 2024 · similarity 0.85)
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.