search

zudi-lin / pytorch_connectomics

PyTorch Connectomics: segmentation toolbox for EM connectomics
http://connectomics.readthedocs.io/
MIT License
169 stars 77 forks source link

Model Input Size and Inference Stride Size in NucMM #125

Closed gitdxj closed 1 year ago

gitdxj commented 1 year ago

In configs/NucMM/NucMM-Mouse-Base.yaml, MODEL.INPUT_SIZE is [33, 97, 97] and INFERENCE.STRIDE is [26, 128, 128] . Doesn't this gap between the model input size and inference stride lead to some pixels uncovered during the test stage?

zudi-lin commented 1 year ago

Hi @gitdxj, thanks for your interest in our work. The inference sample size is controlled by https://github.com/zudi-lin/pytorch_connectomics/blob/master/configs/NucMM/NucMM-Mouse-Base.yaml#L34, so there should not be uncovered pixels. Let me know if you have any questions.

gitdxj commented 1 year ago

Hi @zudi-lin. Your are right. I just checked https://github.com/zudi-lin/pytorch_connectomics/blob/c2c2ab3ae342d3aabd5b474302dee500ef0f6574/connectomics/config/utils.py#L66. It shows the MODEL.INPUT_SIZE is set to be the same as INFERENCE.INPUT_SIZE in inference mode. Thanks for your reply.