Question about instance GT

[tyjiang1997]

Hello, I'd like to ask a question.For the instance split task, how should we generate the corresponding 36 points?

[Duankaiwen]

The GT masks are represented as polygons(a set of points) in the json file, but the number of the points is different for different objects, you need to sample the polygons uniformly to 36 points for all objects.

[tyjiang1997]

The GT masks are represented as polygons(a set of points) in the json file, but the number of the points is different for different objects, you need to sample the polygons uniformly to 36 points for all objects.
Thank you for your reply.I actually want to know if I use my own data, what algorithm should I follow to generate my instance GT

[Duankaiwen]

The GT masks are represented as polygons(a set of points) in the json file, but the number of the points is different for different objects, you need to sample the polygons uniformly to 36 points for all objects.
Thank you for your reply. I actually want to know if I use my own data, what algorithm should I follow to generate my instance GT

The algorithm is on the line 463 in code/mmdet/datasets/pipelines/loading.py

[xiehousen]

The GT masks are represented as polygons(a set of points) in the json file, but the number of the points is different for different objects, you need to sample the polygons uniformly to 36 points for all objects.
Thank you for your reply. I actually want to know if I use my own data, what algorithm should I follow to generate my instance GT

The algorithm is on the line 463 in /code/mmdet/datasets/pipelines/loading.py

Does the instance sem GT have to be Polygons instead of RLE? I tried RLE's and it didn't seem to work well.

[Duankaiwen]

The GT masks are represented as polygons(a set of points) in the json file, but the number of the points is different for different objects, you need to sample the polygons uniformly to 36 points for all objects.
Thank you for your reply. I actually want to know if I use my own data, what algorithm should I follow to generate my instance GT

The algorithm is on the line 463 in /code/mmdet/datasets/pipelines/loading.py

Does the instance sem GT have to be Polygons instead of RLE? I tried RLE's and it didn't seem to work well.

The instance sem GT should be polygons, you may use m = maskUtils.decode(rle) to convert the RLE to the binary masks and use cv.findContours() to convert the binary masks to polygons

[xiehousen]

The GT masks are represented as polygons(a set of points) in the json file, but the number of the points is different for different objects, you need to sample the polygons uniformly to 36 points for all objects.
Thank you for your reply. I actually want to know if I use my own data, what algorithm should I follow to generate my instance GT

The algorithm is on the line 463 in /code/mmdet/datasets/pipelines/loading.py

Does the instance sem GT have to be Polygons instead of RLE? I tried RLE's and it didn't seem to work well.

The instance sem GT should be polygons, you may use m = maskUtils.decode(rle) to convert the RLE to the binary masks and use cv.findContours() to convert the binary masks to polygons

It works. Thanks for your reply.