You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Colours for flat shading are chosen randomly which can make it difficult to distinguish between instances for downstream tasks like generating bounding box. Suggestion is to use separation of colours similar to BlenderProc _colorize_objects_for_instance_segmentation. Requires replacing random_shader() in render.py
The text was updated successfully, but these errors were encountered:
We dont intend for the RGB segmentation visualizations to be used for any downstream tasks. Instead, there should be a corresponding .npy file which has integer class indices for each pixel, which is simpler to parse since you dont have to detect unique colors etc.
Nonetheless thanks a ton for the PRs, I will look into merging all of them soon including the segmentation recoloring as it is still a useful QOL improvement.
I use InstanceSegementation to generate bounding boxes as I need the instances within an object eg the instances from a factory object like CoralFactory. The npy file for InstanceSegmentation is actually the RGB values from the InstanceSegmentation image so there isn't unique integer class indices for the instances (where there is for ObjectSegmentation).
I've found that the colour can vary very slightly across an InstanceSegmentation mask (npy or png file) so more separate colours between instances helps to identify bounding box.
Colours for flat shading are chosen randomly which can make it difficult to distinguish between instances for downstream tasks like generating bounding box. Suggestion is to use separation of colours similar to BlenderProc _colorize_objects_for_instance_segmentation. Requires replacing random_shader() in render.py
The text was updated successfully, but these errors were encountered: