Could not upload Segmentation Mask 1.1 Annotation

[jayhuashi]

My actions before raising this issue

• [x ] Read/searched the docs
• [ x] Searched past issues

I could not upload my Segmentation Mask 1.1 Annotation to the existing task. It says Error: Request failed with status code 400. "Failed to import item ('000015', 'default'): 'Undeclared color (23, 190, 207)

The error is shown below:

I wonder is there a list of pre-defined colors I can use for SegmentationObject mask for the Segmentation Mask 1.1 Format? I have a image with about 3000 instances and each of the mask I use a unique color mapping by the color map "ListedColormap(plt.cm.get_cmap('tab20', num_instances)(range(num_instances)))"

I'm using the web version of the CVAT tool.

[bsekachev]

@jayhuashi

Probably there is not a label with such color rgb(23,190,207) in your task and uploader can not match this color with any label.

[jayhuashi]

@bsekachev

Thank you for the timely response! However, when I inspect the 'Segmentation Mask 1.1' format dataset export from the CVAT, the 'labelmap.txt' records the color corresponding to the class, the 'SegmentationClass' folder is the semantic segmentation map, and the 'SegmentationObject' folder is the instance map which each instance has a unique RGB color. The color for each instance is not declared or defined in any '.txt' file from what I see.

In my case, I followed my inspection to create my annotation and I think this 'Undeclared color (23, 190, 207)' is from what I used for an instance in the 'SegmentationObject' folder, so I'm not sure how I should I resolve this problem.

[bsekachev]

@jayhuashi

How many objects do you have on this frame?

[jayhuashi]

@bsekachev 2847

[bsekachev]

I can suggest that this format can not handle more than 256 objects per image. But let's ask @zhiltsov-max

[bsekachev]

@jayhuashi

It was submitted, that format can not handle more than 256 objects per image. We should improve error message or at least drop a corresponding line to documentation. Please, use another format to import such a big amount of masks.

[zhiltsov-max]

Hi, this format cannot represent more than 256 instances per image. Please consider using some other format if you need more than this number of instances. From mask-specific formats, CamVid also supports 256, and Cityscapes allows up to 1000. KITTI can work in one of its representations with indexed int32 PNG instance masks. I assume, you will need to try something different, such as our CVAT and Datumaro formats or COCO instances, which don't use masks for this data. Another option is OpenImages.

[jayhuashi]

@zhiltsov-max

Thank you for your detailed explanation! I will adjust it based on your suggestion.