I have a pretrained net for semantic segmentation, which has been trained on the cityscapes dataset and its 19 classes (Person, car, traffic sign, …). One of those is "Person". I am only interested in this class.
Is it beneficial to retrain the model on the same dataset for just one class instead of 19?
If yes, what is a good strategy? Keep the weights for the feature extractors and just retrain the last classifying layer(s)? Or tune all the weights?
Thanks in advance!