Closed rohitrango closed 2 years ago
Hello,
actually we skipped segmentation annotations during sampling, hence we havent measure anything. But they should be similar to the box annotations except for the images with crowd
flag set where multiple person boxes have a single segmentation annotation.
We will provide an updated version which also saves segmentation annotations.
thanks for using :)
Hi,
Thank you for the paper and dataset, it is really useful for running ablations faster. I wanted to ask if the segmentations provided in the JSON files are changed in any way from the original
train2017
JSON.Also, do you have any metrics for segmentation?
Thanks a lot.