First timer in image processing - Pardon my cluelessness. Is there a concept of sub labeling in objection identification? I want to label a person and sub label "eye" of a person and train a model to detect if the person's eye is open or closed. i.e do not look for floating eyes in a frame - only look for eyes on a person. Does it improve inference efficiency? How do I do it with LabelImg?
Sounds like face segmentation is what you are looking for. This one looks pretty promising at first glance: https://github.com/zllrunning/face-parsing.PyTorch
Does it improve inference efficiency?
Improve efficiency compared to what? Inference efficiency will probably vary with model size. Compared to vanilla facial recognition, inference will probably be much less efficient because segmentation models are typically larger.
How do I do it with LabelImg?
LabelImg is a tool for annotating image datasets. You can use it to create your own dataset for training/fine-tuning but the use of LabelImg is independent of the segmentation model.