Inquiry about training with custom data

[ToshiEAB]

Hello,

I have a question about training with custom data. After reading this (#70 (comment)), I figured out how to do annotation using Labelme. Now I have a couple questions about annotation for YOLACT.

1. How many annotated images do I need for training? I would like to track a single type of small fish.

2. If I want to track multiple fish that occasionally overlap with each other, should I do annotation using images that have multiple fish overlapping with each other? Or is it enough to do the annotation only with a single fish?

Thanks in advance!