Hi, I like the approach of using flow masks for bacterial cell segmentation. It works really well in highly dense areas. However, I am not getting the expected output. When I pass a labelled image (0 for background and int-n for each cell), I get a 4 dimension image: source, pmap, flow1, flow2. If I understand your documentation, the flow map should encode background as -5, but in my case, it comes as 0. Also, I am not able to reconvert the flow maps into pmaps. Ideally, I would like to train my model to output two channels (pmaps) and then convert to pmap.
Please let me know if you can help; any advice will be highly appreciated.
Hi, I like the approach of using flow masks for bacterial cell segmentation. It works really well in highly dense areas. However, I am not getting the expected output. When I pass a labelled image (0 for background and int-n for each cell), I get a 4 dimension image: source, pmap, flow1, flow2. If I understand your documentation, the flow map should encode background as -5, but in my case, it comes as 0. Also, I am not able to reconvert the flow maps into pmaps. Ideally, I would like to train my model to output two channels (pmaps) and then convert to pmap.
Please let me know if you can help; any advice will be highly appreciated.
This is how I get the flow maps:
And this is how I get the mask back from the flows (channels 2 and 3 from the previous image)