output is plain images

[KoteshwarSurga]

Hi,

I am testing your code on some low resolution images(blur) but the output receiving is plain images.

can u please help me.

[deepak112]

Hi,

Are you using my trained model? Can you tell me what images you are using. And please send me detailed process how you are using my model.

[KoteshwarSurga]

Hi, Thanks for your reply

I am using your model which is in model folder and I run the test.py code using pretrained model. The input images are The output images are so I am getting the plain output.

On Wed, Jun 12, 2019 at 12:28 PM Deepak Birla [email protected] wrote:

Hi,

Are you using my trained model? Can you tell me what images you are using. And please send me detailed process how you are using my model.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/deepak112/Keras-SRGAN/issues/12?email_source=notifications&email_token=AKWA6GMTZAUGDENJPKVKXQDP2CNCZA5CNFSM4HW43TC2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODXPOFEA#issuecomment-501146256, or mute the thread https://github.com/notifications/unsubscribe-auth/AKWA6GKWL4WWPL7YA6QJHETP2CNCZANCNFSM4HW43TCQ .

[souradip-chakraborty]

Hi I am using your pre-trained model. Even if I am changing the image size still it is taking (96,96,3) and giving an error on mismatch.

[mrciolino]

Hi: Trained on 500 images on 320x320 with 4x downsample png files

Using the test.py with the low res input (80x80) I'm getting an entirely white image.

python test.py --input_low_res=../../CompareGANs/shipsnet/Raw/ship/ --output_dir=../../CompareGANs/shipsnet/SRs/agriculture/ships\ SR/ --model_dir='./model/argriculture/gen_model5000.h5' --number_of_images=1000 --test_type='test_lr_images

This is the "best" image I got back:

When using the test.py with the high res input it works fine. python test.py -ihr=../../CompareGANs/skysat/cities/ --output_dir=../../CompareGANs/shipsnet/SRs/agriculture/ships\ SR/ --model_dir='./model/argriculture/gen_model5000.h5' --number_of_images=500 --test_type='test_model'

[mrciolino]

I changed the lr_images (in Utils.py) scale form 4 to 1 since the input wants 80x80 for low res instead of downsampling my already low res 80x80 images to 20x20

def load_test_data(directory, ext, number_of_images = 100):

    files = load_data_from_dirs(load_path(directory), ext)

    if len(files) < number_of_images:
        print("Number of image files are less then you specified")
        print("Please reduce number of images to %d" % len(files))
        sys.exit()
    
    x_test_lr = lr_images(files, 1)
    x_test_lr = normalize(x_test_lr)

return x_test_lr

[mrciolino]

FIxed: On line ~221 the LR images were being "denormalized" when they didn't need to be.

image_batch_lr = denormalize(x_test_lr)

to

image_batch_lr = x_test_lr

Output is now:

[mrciolino]

@KoteshwarSurga I know this is an old thread but did that fix ur problem?

[BassantTolba1234]

Please all, I need the code of implementation this part the part is

{The SRResNet networks were trained with a learning rate of 10−4 and 106 update iterations. We employed the trained MSE-based SRResNet network as initialization for the generator when training the actual GAN to avoid undesired local optima.{

[BassantTolba1234]

Please all, I need the code of implementation this part the part is

{The SRResNet networks were trained with a learning rate of 10−4 and 106 update iterations. We employed the trained MSE-based SRResNet network as initialization for the generator when training the actual GAN to avoid undesired local optima.{