Method Detail: UAV-CrackSeg

Back to Leaderboard
Benchmark: UAV-Crack
Short name: UAV-CrackSeg
Long name: UAV-CrackSeg
Description: Our method, UAV-CrackSeg, aims to improve road crack segmentation in complex backgrounds. 1. Architecture: We use Segformer-b1 as the baseline and introduce the SimAM (Simple Attention Module) parameter-free attention mechanism to enhance feature extraction for fine cracks. 2. Loss Function: A hybrid loss function combining CrossEntropyLoss (1.0) and LovaszLoss (1.0) is designed to optimize boundary connectivity. 3. Strategy: We apply a multi-model weighted fusion strategy to boost the final IoU score. [Model Weights] Baidu Netdisk: https://pan.baidu.com/s/1iDZ_EeWPAi796PCYvY8Wig?pwd=39wn Access Code: 39wn
Reference: N/A
Last submitted: November 16, 2025
Published: November 16, 2025 at 06:01:13
Submissions: 1
Project page / code: https://github.com/hanghang16/UAV-CrackSeg
Open source: Yes

Benchmark performance

Submission Date mIoU (↑) Crack F1 (↑) Crack IoU (↑) Crack Precision (↑) Crack Recall (↑) aAcc (↑)
2025-11-16 06:01 78.7600 75.4400 60.5700 75.0200 75.8700 97.1000