Fix image pre/post processing bug

evaluation.py

@@ -32,7 +32,7 @@ def _eval(trainer, it):
 
         img_c = img.reshape((side, side, 3, image_size, image_size))
         img_c = img_c.transpose(0,1,3,4,2)
-        img_c = img_c*255.0 + 127.5
+        img_c = (img + 1) *127.5
         img_c = np.clip(img_c, 0, 255)
         img_c = img_c.astype(np.uint8)
         img_c = img_c.reshape((side, side, image_size, image_size, 3)).transpose(0,2,1,3,4).reshape((side*image_size, side*image_size, 3))[:,:,::-1]
@@ -43,7 +43,7 @@ def _eval(trainer, it):
         img_t = result.data.get()
         img_t = img_t.reshape( (side, side, 3, image_size, image_size))
         img_t = img_t.transpose(0,1,3,4,2)
-        img_t = img_t*255.0 + 127.5
+        img_t = (img + 1) *127.5
         img_t = np.clip(img_t, 0, 255)
         img_t = img_t.astype(np.uint8)
         img_t = img_t.reshape((side, side, image_size, image_size, 3)).transpose(0,2,1,3,4).reshape((side*image_size, side*image_size, 3))[:,:,::-1]

horse2zebra.py

@@ -40,12 +40,12 @@ def preprocess_image(img):
     #img -= np.array([103.939, 116.779, 123.68], dtype=np.float32)
     """
     img = img.astype("f")
-    img = (img-127.5)/255.0
+    img = img/127.5 - 1
     img = img.transpose((2, 0, 1))
     return img
 
 def postprocess_image(img):
-    img = (img*255.0)+127.5
+    img = (img + 1) *127.5
     img = np.clip(img, 0, 255)
     img = img.astype(np.uint8)
     img.transpose((1, 2, 0))

train.py

@@ -137,7 +137,8 @@ def make_optimizer(model, alpha=0.0002, beta1=0.5):
             'lambda1': args.lambda1,
             'lambda2': args.lambda2,
         #    'lambda3': args.lambda3,
-            'image_size' : args.crop_to
+            'image_size' : args.crop_to,
+            'eval_folder' : args.eval_folder
             #'lambda4': args.lambda4,
         })
 
@@ -157,10 +158,10 @@ def make_optimizer(model, alpha=0.0002, beta1=0.5):
     trainer.extend(extensions.PrintReport(log_keys), trigger=(20, 'iteration'))
     trainer.extend(extensions.ProgressBar(update_interval=50))
 
-    trainer.extend(
-        evaluation(gen_g, gen_f, args.eval_folder, image_size=args.crop_to
-        ), trigger=(args.eval_interval ,'iteration')
-    )
+    #trainer.extend(
+    #    evaluation(gen_g, gen_f, args.eval_folder, image_size=args.crop_to
+    #    ), trigger=(args.eval_interval ,'iteration')
+    #)
 
     # Run the training
     trainer.run()

updater.py

@@ -9,6 +9,8 @@
 from chainer import cuda, optimizers, serializers, Variable
 from chainer import training
 
+from PIL import Image
+
 def cal_l2_sum(h, t):
     return F.sum((h-t)**2)/ np.prod(h.data.shape)
 
@@ -50,6 +52,7 @@ def __init__(self, *args, **kwargs):
         self._lambda2 = params['lambda2']
         #self._lambda3 = params['lambda3']
         self._image_size = params['image_size']
+        self._eval_foler = params['eval_folder']
         #self._lambda4 = params['lambda4']
         self._iter = 0
         self._max_buffer_size = 50
@@ -88,6 +91,16 @@ def getAndUpdateBufferY(self, data):
         id = np.random.randint(0, self._max_buffer_size)
         return self._buffer_y[id, :].reshape((1, 3, self._image_size, self._image_size))
 
+    def save_images(self,img, w=2, h=3):
+        img = cuda.to_cpu(img)
+        img = img.reshape((w, h, 3, self._image_size, self._image_size))
+        img = img.transpose(0,1,3,4,2)
+        img = (img + 1) *127.5
+        img = np.clip(img, 0, 255)
+        img = img.astype(np.uint8)
+        img = img.reshape((w, h, self._image_size, self._image_size, 3)).transpose(0,2,1,3,4).reshape((w*self._image_size, h*self._image_size, 3))[:,:,::-1]
+        Image.fromarray(img).save(self._eval_foler+"/iter_"+str(self._iter)+".jpg")
+
 
     def update_core(self):
         xp = self.gen_g.xp
@@ -176,3 +189,13 @@ def update_core(self):
         chainer.report({'loss_rec': loss_cycle_x}, self.gen_f)
         chainer.report({'loss_adv': loss_gen_g_adv}, self.gen_g)
         chainer.report({'loss_adv': loss_gen_f_adv}, self.gen_f)
+
+        if self._iter%50 ==0:
+            img = xp.zeros((6, 3, w_in, w_in)).astype("f")
+            img[0, : ] = x.data
+            img[1, : ] = x_y.data
+            img[2, : ] = x_y_x.data
+            img[3, : ] = y.data
+            img[4, : ] = y_x.data
+            img[5, : ] = y_x_y.data
+            self.save_images(img)