Research map / Detection and segmentation / CNN features
Frequency Director: Learnable Mixture of Frequency Experts for Unified Concealed Scene Segmentation
Guangqian Guo, Aixi Ren, Xuehui Yu, Pengxu Wei, Yong Guo, shan gao
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 |
|---|---|---|---|
| replace simple adapter with static frequency adapter (ablation) | FreqDirect MiT-B4 · COD10K | S-measure: 0.8172 -> 0.8312 | Improved |
| replace spatial MoE with frequency-domain experts (MoFE) (ablation) | FreqDirect MiT-B4 · COVID-19 | IoU: 0.8230 -> 0.8516 | Improved |
| add Spatial Commonality Anchor to MoFE (ablation) | FreqDirect MiT-B4 · GDD | IoU: 0.8605 -> 0.8938 | Improved |
| add Spatial Commonality Anchor to MoFE (ablation) | FreqDirect MiT-B4 · COD10K | S-measure: 0.8457 -> 0.8542 | Improved |
| use dynamic router instead of fixed uniform routing (ablation) | FreqDirect MiT-B4 · COD10K | S-measure: 0.8344 -> 0.8542 | Improved |
| use 3 experts instead of 2 (ablation) | FreqDirect MiT-B4 · COD10K | S-measure: 0.8419 -> 0.8542 | Improved |
| use 4 experts instead of 3 (ablation) | FreqDirect MiT-B4 · COD10K | S-measure: 0.8542 -> 0.8525 | No clear effect |
| insert FreqDirect in all 4 stages instead of stage 4 only (ablation) | FreqDirect MiT-B4 · COD10K | S-measure: 0.8109 -> 0.8542 | Improved |
| insert in even layers only vs all layers (ablation) | FreqDirect MiT-B4 · COD10K | S-measure: 0.8542 -> 0.8532; params 5.26M -> 4.18M | No clear effect |
| joint training instead of separate training (ablation) | FreqDirect · PlantCAMO zero-shot | S-measure: 0.4198 -> 0.6453 | Improved |
| FreqDirect Swin-L vs FOCUS DINOv2-L | FreqDirect · COVID-19 | IoU: 0.6986 -> 0.8646 | Improved |
Closest papers by meaning
- RUN: Reversible Unfolding Network for Concealed Object Segmentation (ICML 2025 · similarity 0.86)
- Frequency-Adaptive Dilated Convolution for Semantic Segmentation (CVPR 2024 · similarity 0.85)
- Lightweight Frequency Masker for Cross-Domain Few-Shot Semantic Segmentation (NeurIPS 2024 · similarity 0.84)
- FD2-Net: Frequency-Driven Feature Decomposition Network for Infrared-Visible Object Detection (AAAI 2025 · similarity 0.84)
- Frequency Dynamic Convolution for Dense Image Prediction (CVPR 2025 · similarity 0.84)
- LookCloser: Frequency-aware Radiance Field for Tiny-Detail Scene (CVPR 2025 · similarity 0.84)
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.
