Research map / Detection and segmentation / CNN features
Representation Sharing for Fast Object Detector Search and Beyond
Yujie Zhong, Zelu Deng, Sheng Guo, Matthew R. Scott, Weilin Huang
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 |
|---|---|---|---|
| use the proposed diverse transformations instead of DARTS operations (ablation) | FCOS searched head · COCO minival | AP: 39.0 -> 40.3 | Improved |
| add transformations with RFs up to 9 (full set vs RFs {3,5}) (ablation) | searched FCOS head · COCO minival | AP: 39.2 -> 40.3 | Improved |
| allow both standard and depthwise separable conv (vs standard only) (ablation) | searched FCOS head · COCO minival | AP: 39.9 -> 40.3 | Improved |
| allow both conv types (vs depthwise separable only) (ablation) | searched FCOS head · COCO minival | AP: 39.5 -> 40.3 | Improved |
| add decoupling in representation sharing (ablation) | searched FCOS head · COCO minival | AP: 40.0 -> 40.3 | Improved |
| add representation sharing (ablation) | search procedure · COCO minival | search cost (GPU-days): 2.3 -> 0.6 | Improved |
| search on COCO instead of VOC | search procedure · COCO minival | AP: 40.3 -> 40.3 | No clear effect |
| use sequential classification/regression groups instead of parallel groups (ablation) | searched FCOS head · COCO minival | AP: 40.0 -> 40.3 | Improved |
| reverse the order of sequential groups (ablation) | searched FCOS head · COCO minival | AP: drop of 0.1 AP | Hurt |
| replace FCOS head with searched module | FCOS · COCO minival (MobileNetV2) | AP: 31.3 -> 32.7 | Improved |
| replace RetinaNet head with searched module | RetinaNet ResNeXt-101 · COCO minival | AP: 39.8 -> 41.6 | Improved |
| replace mask head with searched cell | Mask R-CNN ResNet-50 · COCO minival | mask AP: 34.2 -> 35.5 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- FCOS — FCOS: Fully Convolutional One-Stage Object Detection (ICCV 2019) · better on 5 of 5 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks (TPAMI 2016 · similarity 0.90)
- NAS-FCOS: Fast Neural Architecture Search for Object Detection (CVPR 2020 · similarity 0.90)
- SM-NAS: Structural-to-Modular Neural Architecture Search for Object Detection (AAAI 2020 · similarity 0.90)
- NAS-FCOS: Efficient Search for Object Detection Architectures (IJCV 2021 · similarity 0.90)
- Auto-FPN: Automatic Network Architecture Adaptation for Object Detection Beyond Classification (ICCV 2019 · similarity 0.90)
- Hit-Detector: Hierarchical Trinity Architecture Search for Object Detection (CVPR 2020 · similarity 0.89)
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.
