fix some bug

fluid/SRCNN/README.md

@@ -78,7 +78,7 @@ mode： 指定模型结构。如果为“base”，使用论文baseline模型结
 ### 预测
 执行以下命令得到模型的预测结果。
 ```
-python infer.py --checkpoint_path="./chkpnt/" --use_gpu=True --image_path="data/val_dataset/set5/baby_GT.bmp"
+python infer.py --checkpoint_path="./chkpnt/" --image_path="data/val_dataset/set5/baby_GT.bmp"
 ```
 需要通过选项`--checkpoint_path`指定模型文件。并使用`--image_path`指定要进行预测的图片。
 

fluid/SRCNN/infer.py

@@ -18,7 +18,7 @@ add_arg = functools.partial(add_arguments, argparser=parser)
 
 add_arg('checkpoint_path',   str,   'model',    "Checkpoint save path.")
 add_arg('image_path',        str,   'data/val_dataset/set5/baby_GT.bmp', "Img data path.")
-add_arg('show_img',          bool,   True,      "show img or not")
+add_arg('show_img',          bool,   False,      "show img or not")
 add_arg('only_reconstruct',  bool,   False,      "If True, input image is seemed as subsampled image")
 add_arg('scale_factor',      int,   3,      "scale factor")
 
@@ -29,8 +29,8 @@ def reconstruct_img(args):
     img_test = cv2.imread(args.image_path)
     yuv_test = cv2.cvtColor(img_test, cv2.COLOR_BGR2YCrCb)       
     img_h, img_w, img_c = img_test.shape
-
-    cv2.imshow('raw image', img_test)
+    if args.show_img:
+    	cv2.imshow('raw image', img_test)
     
     if args.only_reconstruct == False:
         #   blur image and cubic interpolation
@@ -55,16 +55,17 @@ def reconstruct_img(args):
         result_img[result_img >255] = 255
         gap_y = int((img_y.shape[0]-result_img.shape[2])/2)
         gap_x = int((img_y.shape[1]-result_img.shape[3])/2)
-
-        cv2.imshow('input_channel y', img_y)
-
+        if args.show_img:
+            cv2.imshow('input_channel y', img_y)        
+        cv2.imwrite(os.path.join(os.path.split(args.image_path)[0],'beforeSR_'+os.path.split(args.image_path)[1]), img_y)
         img_y[gap_y: gap_y + result_img.shape[2],
                 gap_x: gap_x + result_img.shape[3]]=result_img
+        if args.show_img:
+            cv2.imshow('output_channel y', img_y)
+            cv2.waitKey(0)
 
-        cv2.imshow('output_channel y', img_y)
-        cv2.waitKey(0)
-        
-        cv2.destroyAllWindows()
+            cv2.destroyAllWindows()
+        cv2.imwrite(os.path.join(os.path.split(args.image_path)[0],'afterSR_'+os.path.split(args.image_path)[1]), img_y)
     return img_y
     
 if __name__ == "__main__":

fluid/SRCNN/train.py

@@ -37,17 +37,17 @@ N2=32
 
 def net(X, Y, model_struct):        
     # construct net         
-    conv1 = fluid.layers.nn.conv2d(X, model_struct.n1, model_struct.f1, act='relu', name='conv1' , 
+    conv1 = fluid.layers.conv2d(X, model_struct.n1, model_struct.f1, act='relu', name='conv1' , 
                                 param_attr= fluid.ParamAttr(initializer=fluid.initializer.NormalInitializer(scale=0.001),
                                                         name='conv1_w'),
                                 bias_attr=fluid.ParamAttr(initializer=fluid.initializer.ConstantInitializer(value=0.),
                                                         name='conv1_b'))
-    conv2 = fluid.layers.nn.conv2d(conv1, model_struct.n2, model_struct.f2, act='relu', name='conv2' , 
+    conv2 = fluid.layers.conv2d(conv1, model_struct.n2, model_struct.f2, act='relu', name='conv2' , 
                                 param_attr= fluid.ParamAttr(initializer=fluid.initializer.NormalInitializer(scale=0.001),
                                                         name='conv2_w'),
                                 bias_attr=fluid.ParamAttr(initializer=fluid.initializer.ConstantInitializer(value=0.),
                                                         name='conv2_b'))
-    pred = fluid.layers.nn.conv2d(conv2, 1, model_struct.f3, name='pred', 
+    pred = fluid.layers.conv2d(conv2, 1, model_struct.f3, name='pred', 
                                 param_attr= fluid.ParamAttr(initializer=fluid.initializer.NormalInitializer(scale=0.001),
                                                         name='pred_w'),
                                 bias_attr=fluid.ParamAttr(initializer=fluid.initializer.ConstantInitializer(value=0.),
@@ -94,7 +94,7 @@ def train(args):
                     fluid.framework.default_main_program(),
                     feed=feeder.feed(data),
                     fetch_list=[y_loss])             
-                if batch_id == 0 or backprops_cnt % 100 ==0:   
+                if batch_id == 0:   
                     fluid.io.save_inference_model(args.checkpoint_path, ['image'], [y_predict], exe)  
                     val_loss, val_psnr = validation()
                     print("%i\tEpoch: %d \tCur Cost : %f\t Val Cost: %f\t PSNR :%f" % (backprops_cnt, epoch, np.array(loss[0])[0], val_loss, val_psnr))

fluid/SRCNN/utils.py

@@ -50,8 +50,8 @@ def read_data(data_path, batch_size, ext, scale_factor, bia_size):
                         img_blur = cv2.GaussianBlur(img_patch, (5, 5), 0)
                         img_sumsample = cv2.resize(img_blur, (int(33/scale_factor), int(33/scale_factor)))
                         img_input = cv2.resize(img_blur, (33, 33), interpolation=cv2.INTER_CUBIC)
-                        img_inputs.append(img_input)
-                        img_gths.append(img_gth)
+                        img_inputs.append([img_input])
+                        img_gths.append([img_gth])
                         count_bt += 1
                         if count_bt % batch_size == 0:
                             yield [[np.array(img_inputs), np.array(img_gths)]]