I have a question concerning the training data used for BSDS500 dataset.
In the HED (Holistic edge detection) paper, they used only augmented data from the BSDS500 training and validation dataset; but it seems for RCF, both BSDS data and Pascal data are used for training.
Does RCF still perform better than HED when trained with BSDS500 data only?
We have performed this experiment in the paper. RCF performs better than HED with only BSDS500 dataset for training. Please refer to the paper for details!
Thank you so much for releasing the code!
I have a question concerning the training data used for BSDS500 dataset. In the HED (Holistic edge detection) paper, they used only augmented data from the BSDS500 training and validation dataset; but it seems for RCF, both BSDS data and Pascal data are used for training.
Does RCF still perform better than HED when trained with BSDS500 data only?