Is it possible to train Dataset 2, containing classes B and C, using the weights obtained from training Dataset 1, which includes classes A, B, C, and D, with the resulting best.pt weights?

[MLUSER222]

I attempted this approach to train additional data for classes B and C. However, when I ran detect.py on a video containing classes A, B, C, and D after training Dataset 2, I noticed that only classes B and C from Dataset 2 were being detected. Could this issue be due to the absence of labeling for classes A and D in images from Dataset 2? Or is it best to simply merge Dataset 1 and Dataset 2 and train them together for optimal results?
The data.yaml file is used same format for both Dataset 1 and Dataset 2, and it is formatted as follows:
1: A
2: B
3: C
4: D
both Dataset1 and Dataset2's label files are written with the appropriate indices in the data.yaml files.