search

PJLallen / D2GPLand

Official Implementation of MICCAI 2024 paper "Depth-Driven Geometric Prompt Learning for Laparoscopic Liver Landmark Detection"
MIT License
10 stars 0 forks source link

Difference in evalution #1

Open VincentItaliano opened 1 month ago

VincentItaliano commented 1 month ago

Hello, I evaluated the model with your checkpoints on Google Drive and I found different metric values from those in the paper mIOU = 51.4 (instead of 48.68) and mDice = 65.8 (instead of 59.38)

Do you have any explanation ? Maybe you have uploaded a different checkpoint (better than the one in the paper) from the one you mentioned in the paper

cuiruize commented 1 month ago

Later, we have conducted parameter tuning and re-trained our model to improve the performance of the landmark detection. We have already reverted the Git version to apply all the hyperparameters that we used in our MICCAI work. You may download the code and try the evaluation again. Thanks for your attention.