Method Detail: 123xjl-CrackMix

Back to Leaderboard
Benchmark: UAV-Crack
Short name: 123xjl-CrackMix
Long name: Multi-backbone Crack Segmentation Ensemble
Description: Selected best: 10-model ensemble of grouped/full ResNet34, edge ResNet34, MiT-B2 and ConvNeXt EMA plus grouped scale ResNet34. Full MiT pool averages seeds42/43. Validation-selected family weights; threshold .46; four-flip TTA. Fixed full-data epoch budgets. AI-assisted.
Reference: N/A
Last submitted: September 16, 2026
Published: September 16, 2026 at 18:20:03
Submissions: 11
Project page / code: N/A
Open source: No

Benchmark performance

Submission Date mIoU (↑) Crack F1 (↑) Crack IoU (↑) Crack Precision (↑) Crack Recall (↑) aAcc (↑)
2026-09-16 19:16 82.6200 80.6900 67.6400 80.3600 81.0300 97.7200
2026-09-16 19:12 82.6200 80.6800 67.6200 80.6300 80.7300 97.7300
2026-09-16 19:09 82.6100 80.6800 67.6200 80.3800 80.9900 97.7200
2026-09-16 19:05 82.6200 80.6900 67.6400 80.3600 81.0300 97.7200
2026-09-16 19:01 82.6100 80.6700 67.6000 80.5100 80.8300 97.7200
2026-09-16 18:57 82.6100 80.6800 67.6200 80.3500 81.0200 97.7200
2026-09-16 18:52 82.5100 80.5400 67.4200 80.4500 80.6300 97.7100
2026-09-16 18:41 82.4100 80.4200 67.2600 79.9800 80.8700 97.6900
2026-09-16 18:37 82.3800 80.3800 67.2000 80.0700 80.7000 97.6800
2026-09-16 18:33 82.3700 80.3600 67.1600 80.4900 80.2200 97.6900
2026-09-16 18:20 81.9600 79.8200 66.4100 79.8600 79.7700 97.6300