Research map / Detection and segmentation / CNN features
PKINet-v2: Towards Powerful and Efficient Poly-Kernel Remote Sensing Object Detection
Xinhao Cai, Liulei Li, Gensheng Pei, Zeren Sun, Yazhou Yao, Wenguan Wang
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 hybrid strip+square kernels instead of dense-only (ablation) | PKINet-v2-S · DOTA-v1.0 (ablation, 100-epoch pretraining) | mAP: 78.57 -> 80.11 | Improved |
| use hybrid kernels instead of strip-only (ablation) | PKINet-v2-S · DOTA-v1.0 (ablation, 100-epoch pretraining) | mAP: 79.62 -> 80.11 | Improved |
| use hybrid kernels instead of sparse-only (ablation) | PKINet-v2-S · DOTA-v1.0 (ablation, 100-epoch pretraining) | mAP: 79.36 -> 80.11 | Improved |
| dilations (1,3,3,3) instead of (1,1,1,1) (ablation) | PKINet-v2-S · DOTA-v1.0 (ablation, 100-epoch pretraining) | mAP: 79.88 -> 80.11 | Improved |
| larger dilations (1,4,4,4) vs default (1,3,3,3) (ablation) | PKINet-v2-S · DOTA-v1.0 (ablation, 100-epoch pretraining) | mAP: 80.11 -> 80.01 | Hurt |
| dilations (1,2,2,2) vs default (1,3,3,3) (ablation) | PKINet-v2-S · DOTA-v1.0 (ablation, 100-epoch pretraining) | mAP: 79.98 -> 80.11 | Hurt |
| increase parallel DW branches from 1 to 5 (ablation) | PKINet-v2-S · DOTA-v1.0 (ablation, 100-epoch pretraining) | mAP: 79.62 -> 80.11 | Improved |
| increase parallel DW branches from 4 to 5 (ablation) | PKINet-v2-S · DOTA-v1.0 (ablation, 100-epoch pretraining) | mAP: 79.99 -> 80.11 | Improved |
| apply HKR re-parameterization (ablation) | PKINet-v2-S · DOTA-v1.0 (ablation, 100-epoch pretraining) | FPS: 43.4 -> 54.6 | Improved |
| apply HKR re-parameterization (ablation) | PKINet-v2-T · DOTA-v1.0 (ablation, 100-epoch pretraining) | FPS: 46.2 -> 58.0 | Improved |
| apply HKR re-parameterization (ablation) | PKINet-v2-S · DOTA-v1.0 (ablation, 100-epoch pretraining) | mAP: 80.11 -> 80.11 | No clear effect |
| swap ResNet-50 for PKINet-v2-S backbone | R3Det · DOTA-v1.0 (different detectors) | mAP: 69.70 -> 77.64 | 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.
- ReDet — ReDet: A Rotation-Equivariant Detector for Aerial Object Detection (CVPR 2021) · better on 2 of 2 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
- Poly Kernel Inception Network for Remote Sensing Detection (CVPR 2024 · similarity 0.94)
- SM3Det: A Unified Model for Multi-Modal Remote Sensing Object Detection (AAAI 2026 · similarity 0.89)
- MutDet: Mutually Optimizing Pre-training for Remote Sensing Object Detection (ECCV 2024 · similarity 0.88)
- Strip R-CNN: Large Strip Convolution for Remote Sensing Object Detection (AAAI 2026 · similarity 0.87)
- Benchmarking Object Detectors under Real-World Distribution Shifts in Satellite Imagery (CVPR 2025 · similarity 0.86)
- Alignment-Free RGB-T Salient Object Detection: A Large-Scale Dataset and Progressive Correlation Network (AAAI 2025 · 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.
