Update README.md

zouliangyu · Oct 11, 2017 · df62cdc · df62cdc
1 parent 4b295af
commit df62cdc
Showing 1 changed file with 5 additions and 29 deletions.
diff --git a/DCGAN/README.md b/DCGAN/README.md
@@ -26,37 +26,13 @@ source activate dcgan
 ### Let's start
 We did some slight changes compare to the original code. You can download the [my forked code](https://github.com/layumi/DCGAN-tensorflow) first and then modify the codes as we did.
 
-### 1.Output Size
+### 1.Train
 
-We noticed that directly training on 256x256 input images will lead to random noisy images.
+`python main.py --dataset duke_train --train --input_height 128 --output_height 128 --options 1`
 
-So we use DCGAN to generate 128x128 output then resize it to 256x256 for further training.
-(https://github.com/layumi/Person-reID_GAN/blob/master/DCGAN/utils.py#L48)
+`duke_train` is the dir path which contains images. Here I use the (DukeMTMC-reID)[https://github.com/layumi/DukeMTMC-reID_evaluation] training set. You can change it to your dataset path.
 
-### 2.Random vector
-We use `option=5` to generate images. You can change the range of input random vector.
-(https://github.com/layumi/Person-reID_GAN/blob/master/DCGAN/utils.py#L223)
-
-### 3.Train
-
-Change the visualization setting from `OPTION = 5` to ` OPTION = 1` (https://github.com/layumi/Person-reID_GAN/blob/master/DCGAN/main.py#L100)
-
-Uncomment (https://github.com/layumi/Person-reID_GAN/blob/master/DCGAN/utils.py#L50)
-
-Comment (https://github.com/layumi/Person-reID_GAN/blob/master/DCGAN/utils.py#L48)
-
-Then Train: `python main.py --dataset duke_train --train --input_height 128 --output_height 128`
-
-`duke_train` is the dir path which contains images. Here I use the (DukeMTMC-reID)[https://github.com/layumi/DukeMTMC-reID_evaluation] training set.
-You can change it to your dataset path.
-
-### 4.Test
-Change the visualization setting from `OPTION = 1` to ` OPTION = 5` (https://github.com/layumi/Person-reID_GAN/blob/master/DCGAN/main.py#L100)
-
-Uncomment (https://github.com/layumi/Person-reID_GAN/blob/master/DCGAN/utils.py#L48)
-
-Comment (https://github.com/layumi/Person-reID_GAN/blob/master/DCGAN/utils.py#L50)
-
-Test: `python main.py --dataset duke_train`  
+### 2.Test
 
+`python main.py --dataset duke_train --options 5`  
 It will use your trained model and generate 48,000 images for the following semi-supervised training.