Some questions regarding ~/.labelmerc settings and video data annotation

[STRIVESS]

Dear wkentaro,

Hello! Nice to mmet you. First of all, I would like to give a thumbs-up for the labelme tool developed by you and your team, as it has improved our work efficiency. Recently, I learned to use SegmentAnything for quick semantic segmentation annotation, but I have some questions that I would like to ask you. I hope you can spare some time to help me with them. Thank you!

1. I modified the relevant keyboard shortcuts in the ~/.labelmerc file and restarted labelme, but it seems that the ai-polygon and ai-mask shortcuts are not working (as shown in the image).
   

2. I am currently annotating lawn images, and in the labels.txt file, I have defined two categories: "lawn" and "not_lawn". I want to set the annotation color for the lawn region as green and for the non-lawn region as light red, so that I can identify the boundaries of the lawn during the semantic segmentation training.
   How can I modify the colors for these different labels in the ~/.labelmerc file? Should I change the label_colors to [0, 255, 0]?
   

3. I want to annotate video data. I first split the video into individual frames, and then I use the command labelme data_annotated --labels labels.txt --nodata --keep-prev to annotate each frame. However, I have to manually adjust the translation of the annotation region for each frame. Is there a way where I can annotate a small batch of data and then train a semantic segmentation model using that batch, which can automatically annotate the emaining data?
   Previously, I used the YOLOX algorithm to train over 3000 annotated VOC object detection data and then used the trained model to annotate over 7000 additional data, generating corresponding .xml files. I wrote some blogs YOLOX_Automatic-Labeling and 用yolox算法模型实现自动标注 and I would like to use a similar approach to quickly annotate a large amount of semantic segmentation data.