About erl merging threshold and the challenge timeline

PytorchConnectomics / AxonEM-challenge

utility code for AxonEM challenge

https://axonem.grand-challenge.org/

MIT License

6 stars 1 forks source link

Closed Levishery closed 1 year ago

Levishery commented 1 year ago

Hello,

According to the paper, the original ERL is extended with a “tolerance” threshold. Is it implemented through the parameter merging threshold? If it's true, I found the explanation of this parameter confusing in https://github.com/PytorchConnectomics/AxonEM-challenge/blob/main/challenge_eval/eval_erl.py.

BTW, when will the challenge timeline come out?

Thanks!

donglaiw commented 1 year ago

Thank you for your feedback!

We modify Funke lab's ERL computation with an extra argument merge_thres which is set to 50 to alleviate the effect of skeletonization outliers. https://github.com/PytorchConnectomics/funlib.evaluate/blob/master/funlib/evaluate/run_length.py#L212
Sorry for the delay. We are testing out the docker and it'll be ready in a day or two. will keep you posted