I want to train my model via transfer learning all data classes are custom objects, how many images are required for transfer learning to train any object detection model, each class in numbers? how to get then accuracy better?
It also at some extent depends on the amount of training data used. on a casual note, you should use atleast 0.5-1% of the number of images used during training for each class, while fine tuning the network.
If you have data constraints, try out augmenting the data a bit to meet the required amount.
For this much data, you’ll have to experiment with the number of epochs.