Task 05: Baseline preprocessing code either doesnt extract any patches for some TIFF images or extracts too many for others

Hi,

I'm using the unicorn baseline code from here. While doing sanity checks, it failed for some tiff images

• task-5-case-137.tif: This file is too big (20MB) for a cropout and based on patch preprocessing code in the repo, it dumps too many patches and processing those goes beyond the timelimit
• task-5-case-213.tif: This file is too small and hence patch preprocessing code doesnt get even a single patch for extracting features.

These images identifies potential edge cases that are not handled by the patch preprocessing in the repo. Since this is affecting submission to the Sanity check LB, hence requesting if there would be any update to the code in baseline or competitors are supposed to handle it themselves?

Thanks in advance

Hi @aditya.vartak, thanks for reporting this! We accounted for these edge cases in our challenge baseline, but the changes were not yet propagated to the baseline template. We're currently working on a fix and will update the baseline template repository later today!

Hi, the issue with the challenge baseline template has been resolved! We updated the patch extraction parameters to properly handle small ROIs. Regarding the large ROI (task-5-case-137), we did not encounter issues processing all extracted patches within the time limit using our baseline solutions. If you continue to experience problems with that specific case, feel free to reach out!

Thanks, switching to the challenge baseline from the baseline template helped to resolve such timeout errors

A follow up on this

Getting the following error for another version of the algorithm

The output file 'patch-neural-representation.json' is not valid. JSON does not fulfill schema: instance is too long

Not exactly sure what does it mean. However I ran the same file locally and it worked, with JSON file of size 15MB.

Is this a JSON size issue?

Hi, I'm not entirely sure what's causing the issue, but I'm happy to investigate. Could you let me know the size of the output features and the maximum number of features saved in a file? That might help us pinpoint whether this is related to the size or structure of the output JSON.

Hi, Here is the JSON file generated as output

WSI: task-5-case-137.tif Output feature size: 1152 Tiles processed/features saved: 334

Thanks, I'll investigate this issue and get back to you asap!

The issue has been resolved. The schema for the patch-neural-representation.json previously allowed a maximum feature size of 1024, but we've now increased this limit to 2048. Your algorithm should work as expected with this update. Please let me know if you encounter any further issues!

Thanks for the update.