Feature: custom label colors

[myounus96]

Is this possible?
Like red for person,blue for car,so whenever I annotate a car it should've blue color and whenever I annotate person it should've red color.

[zhiltsov-max]

Currently, there is no way to do this in UI. However, it is possible in Datumaro, where you can provide a custom colormap for export in voc_segmentation format.

[nmanovic]

@m-younus , please don't ignore our issue template. You can choose a color in UI for a label (temporary). I will close the issue. Don't hesitate to re-open if you can share more information how you want it should be implemented.

[zhiltsov-max]

@bsekachev, as we already have a temporary color selector for labels in UI

can we extend this to be persistent and provide same selector at the task page?
Here:

As tasks can have more than N labels, I think it would be nice to provide just a typical color selector like we can see in most of image editors - with exact numbers and a palette.

Example (PDN):

[bsekachev]

@zhiltsov-max

We can, but:

1. First need implement the feature on the server (storing color in the database)
2. I suppose predefined colors not enough in general case. We need something more advanced. But on the first iteration we might support only predefined colors.

[creativesiva]

Currently, there is no way to do this in UI. However, it is possible in Datumaro, where you can provide a custom colormap for export in voc_segmentation format.

Hi Could you please let suggest how to setup custom colors for segmentation using Datumaro
Thanks

[zhiltsov-max]

Please check the last example here.

[creativesiva]

Could you pls help me to fix below issue, I am running this command "datum project import --format voc --input-path /home/aipl/segmentation/images/siva"

my directory structure
+Annotations
+ImageSets
+JPEGImages
+labelmap.txt (changed color codes)

(myannotation) siva@siva-dl:~$ datum project import --format voc --input-path /home/aipl/segmentation/images/siva
2020-08-03 18:25:13,355 WARNING: Failed to import module 'openvino_launcher.py': libpython3.7m.so.1.0: cannot open shared object file: No such file or directory
2020-08-03 18:25:15,691 INFO: Importing project from '/home/aipl/segmentation/images/siva'
2020-08-03 18:25:15,691 INFO: Importing project as 'voc'
2020-08-03 18:25:15,691 INFO: Checking the dataset...
2020-08-03 18:25:15,696 ERROR: Train
Traceback (most recent call last):
File "/home/aipl/anaconda3/envs/myannotation/bin/datum", line 8, in
sys.exit(main())
File "/home/aipl/anaconda3/envs/myannotation/lib/python3.8/site-packages/datumaro/cli/main.py", line 114, in main
return args.command(args)
File "/home/aipl/anaconda3/envs/myannotation/lib/python3.8/site-packages/datumaro/cli/contexts/project/init.py", line 222, in import_command
dataset = project.make_dataset()
File "/home/aipl/anaconda3/envs/myannotation/lib/python3.8/site-packages/datumaro/components/project.py", line 815, in make_dataset
return ProjectDataset(self)
File "/home/aipl/anaconda3/envs/myannotation/lib/python3.8/site-packages/datumaro/components/project.py", line 554, in init
for item in source:
File "/home/aipl/anaconda3/envs/myannotation/lib/python3.8/site-packages/datumaro/plugins/voc_format/extractor.py", line 127, in iter
anns = self._parse_annotations(root_elem)
File "/home/aipl/anaconda3/envs/myannotation/lib/python3.8/site-packages/datumaro/plugins/voc_format/extractor.py", line 147, in _parse_annotations
obj_label_id = self._get_label_id(label_elem.text)
File "/home/aipl/anaconda3/envs/myannotation/lib/python3.8/site-packages/datumaro/plugins/voc_format/extractor.py", line 54, in _get_label_id
assert label_id is not None, label
AssertionError: Train

[zhiltsov-max]

Apparently, the dataset has a Train class, which is not listed in the labelmap.txt, make sure all the classes are listed in this file. Names are case sensitive. More information on the file usage here.

[creativesiva]

It worked now. Thanks a lot.

[creativesiva]

Looks like there is limitation in the #colors, I am trying to add 70 classes (labelmap.txt), But when i download the project after segmentation using UI, the labelmap color codes are changing.