Ground truth/label modification during training (with the data obtained from the



I'm working on an image segmentation algorithm with FCN (Long et al., 2015) as the backbone network.

One idea I have is to use the argmax binary mask obtained from the final score layer (250x250x1) to generate some data (e.g. number of blobs in the mask) to modify the ground truth (e.g. set some pixels in the gt mask to 'ignore' labels) or in some way (partly) extract from the features (similar to RPN layer in FasterRCNN).

Does this violate any deep learning or machine learning rules?


Posted 2018-02-03T18:40:34.860

Reputation: 692

Are you making this modification to labels in the training dataset only (as part of training), or are you also modifying the test dataset labels? – Neil Slater – 2018-02-08T16:41:29.853

No just the training. Test and validation data remain unchanged. – Alex – 2018-02-08T21:58:02.973

So it's modifying the valie of the loss function. Nothing else. – Alex – 2018-02-09T13:39:04.787

No answers