search

xmindflow / deformableLKA

[WACV 2024] Beyond Self-Attention: Deformable Large Kernel Attention for Medical Image Segmentation
https://arxiv.org/abs/2309.00121
157 stars 12 forks source link

I encounter when i reproduce to test 3d synpase #24

Open zyqzyqhappiness opened 2 months ago

zyqzyqhappiness commented 2 months ago

I find the code finally output 13 classes' results, but in inference_synpase.py, you wanna 8 classes like other papers. i wonder is the final result save in "summary.json"? and when i reproduce the code, i find inference_synpase.py didnt work , as well as write in the text. Then my final result is different from your paper, the result is in the image 图片1

Leonngm commented 2 months ago

Hi, for 3D training we use the nnFormer training setup. For Synapse, 13 classes are present but only 8 are evaluated. If you want to consider all 13 classes, compare your results to nnFormer for BTCV. Synapse is a subset of BTCV. The eight classes in Synapse are: Spleen, Right Kidney, Left Kidney, Gallbladder, Liver, Stomach, Aorta and Pancreas.

BTCV has the additional classes: Esophagus, inferior vena cava, portal and splenic veins, right adrenal gland, left adrenal gland.

You can find the result for dice score under "Dice" in the dictionary.