EDD evaluation query

[Shahadate Rezvy]

Hi,

I tried submitting my results in bbox detection and segmentation for the EDD2020 challenge. Could anyone please check whether they are in correct format? Files are attached for one of the test files.

Currently I have the mask overlayed on the image, do I need to submit the binary masks for all the 5 channels for the evaluation to work? Are there any tools or code already available for this conversion?

Thank you

Shahadate

[sali]

Hi,

Your semantic segmentation results should be in .tif format and must include 5 channels. Please see here: https://github.com/sharibox/EndoCV2020/wiki

Related code: https://github.com/sharibox/EDD2020/blob/master/semanticSegmentationMetricEvaluation.py

This should be however straightforward. Just generate masks between [0-1] and concatenate them as 5 channel .tif with each channel representing different class as in the wiki.

Let me know if I can help you more. 

BW,

[sali]

Ok, I have added one more python code to help you combine individual masks to EDD tif format.

- Generate your masks as ImageName_$className.tif, e.g., 00001-test_BE.tif, 00001-test_cancer.tif

- Run this code providing both original image folder and your generated masks, This will generate the 5 channel tif images for you.

code here:

https://github.com/sharibox/EDD2020/blob/master/compositeMasks.py

BW,

[shahad...@gmail.com]

Hi,

I have tried to submit resuts to  EDD2020 two times following the folder structure and zipping the files. However, I had problem submitting my results to the leaderboard each time. Could you please notify me of any known problem?

I have used your code to make the .tif files where 22 images work correctly and I had to manually assemble 2 of them test image 1 and 2 using ImageJ.

Regards

Shahadate

[sali]

Hi,

This should now work. Try again!!!

BW,

SA

[shahad...@gmail.com]

Hi,

I can see some of the values popolated now. As there are no class-wise score, I am assuming my bounding-box formatting might not be correct. Attaching two bounding-box annotations. Appreciate your suggetions on this please.

Regards

Shahadate