Area (Language):

大陆港澳 (简体中文)

台灣 (繁體中文)

 

[Introduction]

 

 

This APP uses SRGAN to generate a High Resolution image from a Low Resolution image to improve the resolution of the image.

The above picture shows an original image (Original HR), after reducing the resolution (Bicubic), and then using SRResNet and SRGAN to increase the image resolution.

 

[Operation steps and instructions]

1. Prepare the dataset

The dataset used by the APP is the dataset of coco2014. Select the dataset to be used in Select Dataset.

If you want to use your own dataset, please press View, copy coco2014, and place it in the same location as coco2014 (press the location opened by View), change it to the folder name you want to set, and replace the image files in train, val and test folders.

 

Note:

Do not delete the "vgg19-dcbb9e9d.pth" file in the model folder.

 

 

2. Prepare data for training

Press 1. Create data lists to generate the json files needed for training.

 

 

3. Train SRResNet

SRResNet must be trained first.

Press 2. train SRResNet to start training.

 

Note:

(1) If you need to resume training, please check the resume mode below and press 2. When train SRResNet, select the model file to continue training. The model file name must be XXX_srresnet.pth.tar.

(2) If you need to set a different batch size, you can fill in the batch size field below.

(3) The trained model is placed in the model folder with a fixed name "checkpoint_srresnet.pth.tar".

(4) Since the training will continue to update "checkpoint_srresnet.pth.tar", if the user needs it, please back up the "checkpoint_srresnet.pth.tar" file by yourself.

 

 

4. train SRGAN

Prerequisite: "checkpoint_srresnet.pth.tar" and "vgg19-dcbb9e9d.pth" must be in the model folder.

Press 3. train SRGAN to start training.

Note:

(1) If you need to continue training, please check the resume mode below and press 3. When train SRGAN, select the model file to continue training. The model file name must be XXX_srgan.pth.tar.

(2) If you need to set a different batch size, you can fill in the batch size field below.

(3) The trained model is placed in the model folder with a fixed name "checkpoint_srgan.pth.tar".

(4) Since the training will continue to update "checkpoint_srgan.pth.tar", if the user needs it, please back up the "checkpoint_srgan.pth.tar" file by yourself.

 

 

5. test

Press "4. super resolve" to select the picture to be tested.

If you need to select other models for testing, please select srresnet model or srgan model in the area below.