Research map / Detection and segmentation / CNN features
Rotation Invariant and Symmetry Aware Pixel Difference Network for Remote Sensing Object Detection
Jialei Zhan, Li Liu, Jiehua Zhang, Yuhang Xie, Yongxiang Liu, Jiangming Chen, Ming-Ming Cheng
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 RIS-PDC to LSKNet-S backbone (single-scale) | LSKNet-S · DOTA-v1.0 test (single-scale) | mAP: 77.49 -> 78.53 | Improved |
| add RIS-PDC to LSKNet-S backbone (multi-scale) | LSKNet-S · DOTA-v1.0 test (multi-scale) | mAP: 81.64 -> 81.81 | Improved |
| add RIS-PDC to LSKNet-T backbone | LSKNet-T · DOTA-v1.0 test (single-scale) | mAP: 74.83 -> 76.92 | Improved |
| add RIS-PDC to LSKNet-S, increases FLOPs | LSKNet-S · DOTA-v1.0 test | FLOPs: 161G -> 206G | Hurt |
| add RIS-PDC to LSKNet-S on HRSC2016 | LSKNet-S · HRSC2016 | mAP(12): 98.46 -> 98.60 | Improved |
| add RIS-PDC to LSKNet-S on DIOR-R | LSKNet-S · DIOR-R | mAP: 65.90 -> 67.28 | Improved |
| use RIS-PDC at layer1 and head but not layer2 (ablation) | RIS-PiDiNet-S · DOTA-v1.0 test (single-scale, ablation) | mAP: 77.75 -> 77.60 (FLOPs 206G -> 164G) | Mixed |
| use learnable harmonic orders instead of fixed three orders (ablation) | S-PDC · DOTA-v1.0 test (single-scale, ablation) | mAP: 77.23 -> 77.75 | Improved |
| add polar coordinate terms to learned harmonic kernel (ablation) | S-PDC · DOTA-v1.0 test (single-scale, ablation) | mAP: 77.31 -> 77.75 | Improved |
| increase learnable harmonic orders to five (ablation) | S-PDC · DOTA-v1.0 test (single-scale, ablation) | mAP: 77.75 -> 77.48 | Hurt |
| use S-PDC instead of Laplacian, Gabor, steerable filters or Hu/Zernike moments (ablation) | RIS-PDC · DOTA-v1.0 test (single-scale, ablation) | mAP: 77.05 -> 77.75 (best alternative: steerable filters) | Improved |
| use 8 SO(2) rotation directions rather than 2, 4 or 16 (ablation) | RIS-PDC · DOTA-v1.0 test (single-scale, ablation) | mAP | Improved |
Closest papers by meaning
- ReDet: A Rotation-Equivariant Detector for Aerial Object Detection (CVPR 2021 · similarity 0.90)
- Weakly Supervised Rotation-Invariant Aerial Object Detection Network (CVPR 2022 · similarity 0.89)
- FRED: Towards a Full Rotation-Equivariance in Aerial Image Object Detection (AAAI 2024 · similarity 0.88)
- Measuring the Impact of Rotation Equivariance on Aerial Object Detection (ICCV 2025 · similarity 0.88)
- R2Det: Exploring Relaxed Rotation Equivariance in 2D Object Detection (ICLR 2025 · similarity 0.87)
- Leveraging 3D Geometric Priors in 2D Rotation Symmetry Detection (CVPR 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.
