Im using "COCO-InstanceSegmentation/mask_rcnn_X_101_32x8d_FPN_3x.yaml" and gave a custom dataset that holds picture of orange, apple, and pineapple. Now my model is ready and is able to predict all 3 fruits. Let's say now I need to add more fruits or add more data for pineapple (which I didn't feed enough to the system already).
Can I load the model and start the training with a new dataset(on top of weights which it already has)? and how good the result will be?
will it affect the previous predictions? Since orange and apple are predicted correctly and I don't want to mess that up. I'm going to send in pictures of pineapple only in this second training and also planning to add new classes like Banana or Watermelon in later training.
Is this the right way to do it? or Let me know if we have better options
