Hi. My team is working on a project focused on one of the animals in the Model Zoo, and I’m trying to see if we could use the pretrained data from the Model Zoo. However, there are a few differences between what’s available and what we need, and I want to know if there’s a way to modify it to meet our needs.
First, is it possible to use if for a multi animal project? The currently available model seems to be a single animal project, but we need to study interactions between multiple animals at once. I know it’s possible to convert a normal single animal project to multi animal by modifying the config file and using deeplabcut.convert2_maDLC(), but so far I haven’t had any luck with it.
Second, the Model Zoo project uses fewer body parts than we do, and some of them are slightly different (ankle vs. foot for example). From what I’ve read, we should be able to add new body parts and remove others by modifying the config file. However, in most cases the different body parts are only slightly different, so while I doubt there’s an easy way to do so, if it’s even possible, I feel I should ask: is it possible to use the weights for the ankle as a starting point for training the network to recognize the foot? On a related note, some of the body parts that are the same across projects have slightly different names (for example a number before the text name in one project but not in the other). Would altering the names so they match cause problems (I’m kind of assuming it would)?
Finally, the Model Zoo project is set to run with a resnet50, but we have been using a different network type so far. Is it possible to modify the project to use a different network?
Thank you for any help you can provide!