About MVS annotation in COCO format

[xingyizhou]

Hi @pkontschieder , By following the instruction, it seems that the coco-format json file used here is missing for Mapillary Vista. Is this the same one created in this repo?

Also, the extracting data script assumes the MVS dataset version is v1.1, while the most recent version I requested if v1.2. Can I assume I can use v1.2 for the challenge since there is no much difference between the versions?

[ozendelait]

@xingyizhou : until we get a confirmation, I have added the prepare_vista script as a step during the extractor step. Please test this script: https://github.com/ozendelait/rvc_devkit/blob/master/segmentation/convert_mvd_coco.sh I have renamed all "mvs" abbreviations to the correct "mvd". All you have to do is rename the mvs folder, everthing else stay the same

[ozendelait]

Oh, and I also updated the extractor to use v1.2 instead of v1.1 although both should work with the converter script and both are fine for usage during training (the newer v1.2 is the prefered one)

[pkontschieder]

The difference between v1.1 and v1.2 is that RGB images now have privacy blurring. There is no difference wrt data loading, filenames, etc. @xingyizhou @ozendelait

[ozendelait]

Thanks for the clarification. Issue closed