The organizer indicates that submission file has
(10400, 256, 256) (# of files, w, h) numpy array for Phase 1 (validation files sequence).
I estimated labels and created the numpy array (.npy).
I submitted the file which was compressed (zip) to the site.
However, the result showed "no score (---)".
Could you tell me the format of the submission file, or how the submission file is created.
Thanks,
Posted by: Skawakita @ April 17, 2021, 5:57 a.m.I cleared my question by myself.
No score(---) means "Calculating IoU".
sorry to bother you.
Best,
Quick question, is the size of the submission (# tiles, w, h) or (# tiles, h, w ) ?
I know in this case the numbers are the same, since both w and h are 256. But it is not the same to upload images with the width as the first or second dimenion. Bu default numpy loads the height of in image as the first dimension. If submissions are required as (# tiles, w, h), do we have to transpose the array before submiting ?
I tried to upload one submission with both shapes but got the same result anyway, which is strange.
Posted by: jsensio @ April 17, 2021, 2:59 p.m.From the site, (# of files, w, h) is correct.
https://nasa-impact.github.io/etci2021/
The segmentation results (prediction labels) will be submitted to the CodaLab competition website for evaluation as:
(10400, 256, 256) (# of files, w, h) numpy array for Phase 1 (validation files sequence), and
(12348, 256, 256) (# of files, w, h) for Phase 2 (test files sequence) respectively.
It would be good if it could solve your issue.
Best.
Posted by: Skawakita @ April 17, 2021, 11:51 p.m.Just to clarify the flood class in the mask shall be set to 1, right?
Posted by: hpysr_s @ April 19, 2021, 7:43 p.m.Yes, that is correct.
Did you by chance find out the reason why the same results are returned even when you change the shape of the submission array? We are experiencing the same effect which seems strange.
Posted by: mwlan @ April 26, 2021, 7:20 p.m.