Research map / Detection and segmentation / CNN features
Aggregation With Feature Detection
Shuyang Sun, Xiaoyu Yue, Xiaojuan Qi, Wanli Ouyang, Victor Adrian Prisacariu, Philip H.S. Torr
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 DEPLA module | ResNet-50 · ImageNet-1k val (standard training) | top-1 accuracy: 76.9 -> 78.6 (+1.6) | Improved |
| add DEPLA module | SEResNet-50 · ImageNet-1k val (standard training) | top-1 accuracy: 77.5 -> 79.1 | Improved |
| add DEPLA module plus better training strategy | ResNet-50 · ImageNet-1k val | top-1 accuracy: 76.2 -> 80.6 | Improved |
| add DEPLA to well-tuned SEResNet-50++ | SEResNet-50++ · ImageNet-1k val | top-1 accuracy: 79.4 -> 81.3 | Improved |
| add DEPLA module | SEResNet-101 · ImageNet-1k val | top-1 accuracy: 77.7 -> 82.1 | Improved |
| use adaptive BFN instead of plain BFN (ablation) | ResNet-50-DEPLA · ImageNet-1k val | top-1 accuracy: 78.6 -> 78.8 | Improved |
| replace FPN with FPN-DEPLA | Mask R-CNN ResNet50 · COCO val2017 | box AP: 38.0 -> 39.0 | Improved |
| lower softmax temperature T from 1.0 to 0.2 (ablation) | FPN-DEPLA · COCO val2017 | small-object AP: 23.5 -> 23.8 | Improved |
| vary softmax temperature T (ablation) | FPN-DEPLA · COCO val2017 | box AP: 38.9 / 39.1 / 39.0 | No clear effect |
| add DEPLA to FishNet150 backbone | Faster R-CNN FPN · COCO val2017 | box AP: 40.6 -> 43.1 | Improved |
| add DEPLA to FishNet150 backbone | Mask R-CNN · COCO val2017 | mask AP: 37.0 -> 39.4 | Improved |
| add DEPLA to FishNet150 backbone | Mask R-CNN · COCO val2017 | box AP: 41.5 -> 44.1 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- FishNet150 — FishNet: A Versatile Backbone for Image, Region, and Pixel Level Prediction (NeurIPS 2018) · better on 3 of 3 shared results
- Res2Net-101 — Res2Net: A New Multi-Scale Backbone Architecture (TPAMI 2019) · better on 1 of 1 shared result
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
- Deep Layer Aggregation (CVPR 2018 · similarity 0.86)
- Deformable Part Region Learning and Feature Aggregation Tree Representation for Object Detection (TPAMI 2023 · similarity 0.84)
- Flow-Guided Feature Aggregation for Video Object Detection (ICCV 2017 · similarity 0.84)
- Container: Context Aggregation Networks (NeurIPS 2021 · similarity 0.84)
- Unifying Feature and Cost Aggregation with Transformers for Semantic and Visual Correspondence (ICLR 2024 · similarity 0.83)
- Towards Implicit Aggregation: Robust Image Representation for Place Recognition in the Transformer Era (NeurIPS 2025 · 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.
