Wenny

yes, I also replaced the pre-trained "prototxt" file with the training "prototxt" file, but another problem appeared，training pipeline can't output the colored images

I replaced the corresponding part in the colorize.py with ./models/colorization_train_val_v2.prototxt and colornet_iter_135000.caffemodel(caffemodel saved in training process). but the output image is from net.blobs['class8_ab'].data in colorize.py. this 'class8_ab' blobs is not...

so you use "./models/**colorization_deploy_v2.prototxt**" and the model you train(**not** default='./models/colorization_release_v2.caffemodel') and you get the right result？i just hope that the colored images I get are trained on the model with...

I'm sorry to bother you again，I don't know if there was a problem with the pre-processing data in training process（I processed my images dataset into **LMDB files，BGR images**）,I ran the...

thanks，I read your script. it seems that you did not preprocess the training data into BGR images，so this can directly get the correct result.

oh! yes, thanks a lot！