junxnone / aiwiki

AI Wiki
https://junxnone.github.io/aiwiki
18 stars 2 forks source link

paper MuSC #467

Open junxnone opened 4 months ago

junxnone commented 4 months ago

MuSC

Arch

Image

LNAMD

MSM

RsCIN

Inference

Performance

Backbones Pre-training image size AUROC-cls F1-max-cls AP-cls AUROC-segm F1-max-segm AP-segm PRO-segm Inference times(ms/image)
ViT-B-32 CLIP 256 87.99 92.31 94.38 93.08 42.06 37.21 72.62 48.33
dino_vitbase16 DINO 256 89.39 93.77 95.37 95.83 54.02 52.84 84.24 85.97
ViT-B-32 CLIP 512 89.91 92.72 95.12 95.73 53.32 52.33 83.72 95.74
ViT-B-16 CLIP 256 92.78 93.98 96.59 96.21 52.48 50.23 87 86.68
dino_vitbase16 DINO 512 94.11 96.13 97.26 97.78 62.07 63.2 92.49 458.5
ViT-B-16 CLIP 512 94.2 95.2 97.34 97.09 61.24 61.45 91.67 450.5
ViT-B-16-plus-240 CLIP 240 94.77 95.43 97.6 96.26 52.23 50.27 87.7 85.25
dinov2_vitb14 DINO_v2 336 95.67 96.8 97.95 97.74 60.23 59.45 93.84 209.1
ViT-B-16-plus-240 CLIP 512 95.69 96.5 98.11 97.28 60.71 61.29 92.14 506.4
ViT-L-14 CLIP 518 95.94 96.32 98.3 97.42 63.06 63.67 92.92 933.3
ViT-L-14 CLIP 336 96.06 96.65 98.25 97.24 59.41 58.1 91.69 266
dinov2_vitb14 DINO_v2 518 96.31 96.87 98.32 98.07 64.65 65.31 95.59 755
ViT-L-14-336 CLIP 336 96.4 96.44 98.3 97.03 57.51 55.44 92.18 270.2
dinov2_vitl14 DINO_v2 336 96.84 97.45 98.68 98.17 61.77 61.21 94.62 281.4
dinov2_vitl14 DINO_v2 518 97.08 97.13 98.82 98.34 66.15 67.39 96.16 1015.1
ViT-L-14-336 CLIP 518 97.77 97.37 99.07 97.11 62.16 62.26 93.45 955.3

Q?