Research map / Detection and segmentation / CNN features
DEFANet: Dual-Path Edge-Target Collaboration with Frequency-Aware Enhancement for Infrared Small Target Detection
Shuaiyuan Du, Yang Xiao, Zhiguo Cao
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 auxiliary edge branch supervision (Easy ECT) (ablation) | baseline UNet · mean F1 over MDvsFA, LAFT, SIRST | mean F1: 0.483 -> 0.496 (+2.7%) | Improved |
| inject edge features into main branch with simple attention (Easy EGI) (ablation) | baseline + Easy ECT · mean F1 over MDvsFA, LAFT, SIRST | mean F1: 0.496 -> 0.521 | Improved |
| use dual-path group-wise EGI instead of Easy EGI (ablation) | baseline · mean F1 over MDvsFA, LAFT, SIRST | mean F1: 0.521 -> 0.580 | Improved |
| add FADE frequency-aware enhancement to main branch only (ablation) | baseline · mean F1 over MDvsFA, LAFT, SIRST | mean F1: 0.483 -> 0.552 | Improved |
| add both EGI and FADE (ablation) | baseline · mean F1 over MDvsFA, LAFT, SIRST | mean F1: 0.483 -> 0.635 | Improved |
| feed low-frequency features only to edge branch (ablation) | DEFANet w/o FADE · mean F1 over MDvsFA, LAFT, SIRST | mean F1: 0.580 -> 0.521 | Hurt |
| feed high-frequency features to edge branch (ablation) | DEFANet w/o FADE · mean F1 over MDvsFA, LAFT, SIRST | mean F1: 0.580 -> 0.600 | Improved |
| feed both frequencies to edge branch (ablation) | DEFANet w/o FADE · mean F1 over MDvsFA, LAFT, SIRST | mean F1: 0.580 -> 0.589 | No clear effect |
| main branch low-frequency only (ablation) | DEFANet w/o FADE · mean F1 over MDvsFA, LAFT, SIRST | mean F1: 0.580 -> 0.579 | No clear effect |
| insert EGI at layer 5 (ablation) | DEFANet w/o EGI · mean F1 over MDvsFA, LAFT, SIRST | mean F1: 0.552 -> 0.549 | Hurt |
| insert EGI at layers 2 and 3 instead of all layers (ablation) | DEFANet · mean F1 over MDvsFA, LAFT, SIRST | mean F1: 0.637 (all layers) vs 0.635 (layers 2,3) | No clear effect |
| replace edge features in EGI with zeros (ablation) | DEFANet · mean F1 over MDvsFA, LAFT, SIRST | mean F1: 0.635 -> 0.491 | Hurt |
Closest papers by meaning
- Rethinking Evaluation of Infrared Small Target Detection (NeurIPS 2025 · similarity 0.87)
- Rethinking Infrared Small Target Detection: A Foundation Driven Efficient Paradigm (NeurIPS 2026 · similarity 0.87)
- IRPruneDet: Efficient Infrared Small Target Detection via Wavelet Structure-Regularized Soft Channel Pruning (AAAI 2024 · similarity 0.85)
- Semi-supervised Infrared Small Target Detection with Thermodynamic-Inspired Uneven Perturbation and Confidence Adaptation (AAAI 2025 · similarity 0.85)
- Pinwheel-shaped Convolution and Scale-based Dynamic Loss for Infrared Small Target Detection (AAAI 2025 · similarity 0.84)
- FD2-Net: Frequency-Driven Feature Decomposition Network for Infrared-Visible Object Detection (AAAI 2025 · similarity 0.83)
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.
