Landmark Detection on DHA 25-shot
1.53MRECDPM-align (Ours, λ=5)
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| CDPM-align (Ours, λ=5)Shot count=25-shot, Alignment weight (lambda)=52026.06 | 1.53 | 0.97 | 91.13 | 2.77 | |
| CDPM (NIH, λ=0)Shot count=25-shot, Pre-training=NIH, Alignment weight (lambda)=02026.06 | 1.62 | 1.19 | 89.8 | 3.18 | |
| Di Via et al.Shot count=25-shot2026.06 | 2.22 | 1.48 | 88.28 | 3.67 | |
| Zhu et al.Shot count=25-shot2026.06 | 2.87 | 4.95 | 83.13 | 4.04 | |
| ResNet-101 (ImageNet)Shot count=25-shot, Backbone=ResNet-101, Pre-training=ImageNet2026.06 | 3.95 | 4.39 | 84.07 | 7.65 | |
| CDPM ScratchShot count=25-shot, Pre-training=Scratch2026.06 | 4.22 | 2.96 | 84.5 | 14.43 | |
| ResNet-101 (MoCo v3)Shot count=25-shot, Backbone=ResNet-101, Pre-training=MoCo v32026.06 | 9.28 | 9.64 | 64.5 | 45.29 | |
| ResNet-101 (SimCLR v2)Shot count=25-shot, Backbone=ResNet-101, Pre-training=SimCLR v22026.06 | 9.47 | 9.82 | 64.97 | 46.99 | |
| ResNet-101 (DINO)Shot count=25-shot, Backbone=ResNet-101, Pre-training=DINO2026.06 | 10.16 | 10.24 | 63.83 | 46.98 |