Research map / Detection and segmentation / CNN features
PackDet: Packed Long-Head Object Detector
Kun Ding, Guojin He, Huxiang Gu, Zisha Zhong, Shiming Xiang, Chunhong Pan
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 CH-GN cross-head group normalization (no PackOp) (ablation) | PackDet head, Me=0 · COCO minival | AP: 38.9 -> 39.5 | Improved |
| add CH-GN cross-head group normalization (no PackOp) (ablation) | PackDet head, Me=3 · COCO minival | AP: 39.4 -> 39.7 | Improved |
| apply packing at test time only (ablation) | multi-branch head, Me=0 · COCO minival | AP: 38.9 -> 22.9 | Hurt |
| train and test with PackOp plus CH-GN (ablation) | PackDet, Me=0 · COCO minival | inference time (ms): 67.4 -> 64.4 | Improved |
| train and test with PackOp plus CH-GN (ablation) | PackDet, Me=3 · COCO minival | inference time (ms): 75.0 -> 65.3 | Improved |
| train and test with PackOp plus CH-GN (ablation) | PackDet, Me=3 · COCO minival | AP: 39.4 -> 40.1 | Improved |
| pack three extra feature maps (Me=3 instead of 0) (ablation) | PackDet · COCO minival | AP: 39.7 -> 40.1 | Improved |
| pack P3-P7 with an alternative pattern instead of the default (ablation) | PackDet head · COCO minival | AP: 39.7 -> 39.3 | Hurt |
| use stair pattern instead of grid pattern for P6 packing (ablation) | PackDet head · COCO minival, P6 | AP: 4.1 -> 12.6 | Improved |
| use stair pattern instead of side-by-side for P6 packing (ablation) | PackDet head · COCO minival, P6 | AP: 9.3 -> 12.6 | Improved |
| pack P5-P7 with PackOp (ablation) | PackDet head · COCO minival, P5-P7 | AP: 24.4 -> 24.7 | Improved |
| pack same-scale P5 features with grid pattern (ablation) | PackDet head · COCO minival, P5 | AP: 14.7 -> 14.1 | Hurt |
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.
- FCOS — FCOS: Fully Convolutional One-Stage Object Detection (ICCV 2019) · better on 8 of 8 shared results
- Cascade R-CNN — Cascade R-CNN: Delving Into High Quality Object Detection (CVPR 2018) · 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
- CrossDet: Crossline Representation for Object Detection (ICCV 2021 · similarity 0.81)
- BorderDet: Border Feature for Dense Object Detection (ECCV 2020 · similarity 0.79)
- Precise Detection in Densely Packed Scenes (CVPR 2019 · similarity 0.79)
- Theoretically Achieving Continuous Representation of Oriented Bounding Boxes (CVPR 2024 · similarity 0.78)
- COUNTS: Benchmarking Object Detectors and Multimodal Large Language Models under Distribution Shifts (CVPR 2025 · similarity 0.78)
- Bounding Box Stability against Feature Dropout Reflects Detector Generalization across Environments (ICLR 2024 · similarity 0.78)
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.
