how to use my own training images?

[twinsyssy1018]

hi, i want use some super-resolution images to training srgan ,but i can not find the way to change image dir . could you provide this?

[buriburisuri]

@twinsyssy1018 Sorry for late reply. ( enjoyed long holidays )

If you want to use your own dataset, you can do that as same way as other normal tensorflow jobs.( feed dictionary or queue ... ). Please read https://www.tensorflow.org/how_tos/reading_data/ . Thank your.

[twinsyssy1018]

@buriburisuri thank you very much

[xolott]

HI,

I'm trying to train this model with the LFW dataset. So far, I made the loading code of the entire dataset and convert it to a tensors list. But when I pass this images through the generator network

    # generator network
    with tf.sg_context(name='generator', act='relu', bn=True):
    gen = (x_small
           .sg_conv(dim=32)
           .sg_conv()
           .sg_conv(dim=4, act='sigmoid', bn=False)
           .sg_periodic_shuffle(factor=2))

It gave me a (1333,128,160,1) output tensor, from a (1333,128,160,3) input tensor. So when the code concat the x_nearest with x and with gen, it give me 2 differents shape tensors of (1333,128,160,4) and (1333,128,160,6)

I don't understand how to change this network to fit this data set. This is a difficult subject to me.

[ratsuki]

@xolott I get the same problem when I use my own RGB images,it give me two differents tensors of (32,28,28,6) and (32,28,28,4)
can you resolve this problem?I will be grateful to you.