I modify the raw code of pytorch v1.0 version to add validation loss computing capacity, here are my modifications:
first, i add ground truth labels in roibatchLoader:
then, i replace if self.training with if True in two files to compute both train and validation loss:
one in faster_rcnn.py:
another in prn.py:
When above modifications are made, i test the model trained with vgg16 and i get map of about 70% which is higher than origin inal about 65%, so here is my question, why does this happen?
I modify the raw code of pytorch v1.0 version to add validation loss computing capacity, here are my modifications: first, i add ground truth labels in roibatchLoader:
then, i replace if self.training with if True in two files to compute both train and validation loss:
one in faster_rcnn.py:
another in prn.py:
When above modifications are made, i test the model trained with vgg16 and i get map of about 70% which is higher than origin inal about 65%, so here is my question, why does this happen?