Refine code (PaddlePaddle#44)

* refine code * test * fix apps * update readme * rm unused code * fix apps output when input is image * clean code * update requirements.txt
AnnaTrainingG · Oct 22, 2020 · 3e4a59f · 3e4a59f
1 parent 359db9c
commit 3e4a59f
Show file tree

Hide file tree

Showing 45 changed files with 560 additions and 1,014 deletions.
diff --git a/README.md b/README.md
@@ -61,7 +61,7 @@ pip install -v -e .  # or "python setup.py develop"
 Please refer to [data prepare](./docs/data_prepare.md) for dataset preparation.
 
 ## Get Start
-Please refer [get stated](./docs/get_started.md) for the basic usage of PaddleGAN.
+Please refer [get started](./docs/get_started.md) for the basic usage of PaddleGAN.
 
 ## Model tutorial
 * [Pixel2Pixel and CycleGAN](./docs/tutorials/pix2pix_cyclegan.md)

diff --git a/configs/cyclegan_cityscapes.yaml b/configs/cyclegan_cityscapes.yaml
@@ -1,5 +1,4 @@
 epochs: 200
-isTrain: True
 output_dir: output_dir
 lambda_A: 10.0
 lambda_B: 10.0
@@ -39,12 +38,12 @@ dataset:
     transforms:
       - name: Resize
         size: [286, 286]
-        interpolation: 2 #cv2.INTER_CUBIC
+        interpolation: 'bicubic' #cv2.INTER_CUBIC
       - name: RandomCrop
-        output_size: [256, 256]
+        size: [256, 256]
       - name: RandomHorizontalFlip
         prob: 0.5
-      - name: Permute
+      - name: Transpose
       - name: Normalize
         mean: [127.5, 127.5, 127.5]
         std: [127.5, 127.5, 127.5]
@@ -60,8 +59,8 @@ dataset:
     transforms:
       - name: Resize
         size: [256, 256]
-        interpolation: 2 #cv2.INTER_CUBIC
-      - name: Permute
+        interpolation: 'bicubic' #cv2.INTER_CUBIC
+      - name: Transpose
       - name: Normalize
         mean: [127.5, 127.5, 127.5]
         std: [127.5, 127.5, 127.5]

diff --git a/configs/cyclegan_horse2zebra.yaml b/configs/cyclegan_horse2zebra.yaml
@@ -1,5 +1,4 @@
 epochs: 200
-isTrain: True
 output_dir: output_dir
 lambda_A: 10.0
 lambda_B: 10.0
@@ -38,12 +37,12 @@ dataset:
     transforms:
       - name: Resize
         size: [286, 286]
-        interpolation: 2 #cv2.INTER_CUBIC
+        interpolation: 'bicubic' #cv2.INTER_CUBIC
       - name: RandomCrop
         output_size: [256, 256]
       - name: RandomHorizontalFlip
         prob: 0.5
-      - name: Permute
+      - name: Transpose
       - name: Normalize
         mean: [127.5, 127.5, 127.5]
         std: [127.5, 127.5, 127.5]
@@ -60,8 +59,8 @@ dataset:
       transform:
       - name: Resize
         size: [256, 256]
-        interpolation: 2 #cv2.INTER_CUBIC
-      - name: Permute
+        interpolation: 'bicubic' #cv2.INTER_CUBIC
+      - name: Transpose
       - name: Normalize
         mean: [127.5, 127.5, 127.5]
         std: [127.5, 127.5, 127.5]

diff --git a/configs/makeup.yaml b/configs/makeup.yaml
@@ -1,5 +1,4 @@
 epochs: 100
-isTrain: True
 output_dir: tmp
 checkpoints_dir: checkpoints
 lambda_A: 10.0
@@ -24,14 +23,14 @@ dataset:
   train:
     name: MakeupDataset
     trans_size: 256
-    dataroot: MT-Dataset
+    dataroot: data/MT-Dataset
     cls_list: [non-makeup, makeup]
     phase: train
     pool_size: 16
   test:
     name: MakeupDataset
     trans_size: 256
-    dataroot: MT-Dataset
+    dataroot: data/MT-Dataset
     cls_list: [non-makeup, makeup]
     phase: test
     pool_size: 16

diff --git a/configs/pix2pix_cityscapes.yaml b/configs/pix2pix_cityscapes.yaml
@@ -1,5 +1,4 @@
 epochs: 200
-isTrain: True
 output_dir: output_dir
 lambda_L1: 100
 
@@ -36,15 +35,15 @@ dataset:
     transforms:
       - name: Resize
         size: [286, 286]
-        interpolation: 2 #cv2.INTER_CUBIC
+        interpolation: 'bicubic' #cv2.INTER_CUBIC
         keys: [image, image]
       - name: PairedRandomCrop
-        output_size: [256, 256]
+        size: [256, 256]
         keys: [image, image]
       - name: PairedRandomHorizontalFlip
         prob: 0.5
         keys: [image, image]
-      - name: Permute
+      - name: Transpose
         keys: [image, image]
       - name: Normalize
         mean: [127.5, 127.5, 127.5]
@@ -63,9 +62,9 @@ dataset:
     transforms:
       - name: Resize
         size: [256, 256]
-        interpolation: 2 #cv2.INTER_CUBIC
+        interpolation: 'bicubic' #cv2.INTER_CUBIC
         keys: [image, image]
-      - name: Permute
+      - name: Transpose
         keys: [image, image]
       - name: Normalize
         mean: [127.5, 127.5, 127.5]

diff --git a/configs/pix2pix_cityscapes_2gpus.yaml b/configs/pix2pix_cityscapes_2gpus.yaml
@@ -1,5 +1,4 @@
 epochs: 200
-isTrain: True
 output_dir: output_dir
 lambda_L1: 100
 
@@ -35,15 +34,15 @@ dataset:
     transforms:
       - name: Resize
         size: [286, 286]
-        interpolation: 2 #cv2.INTER_CUBIC
+        interpolation: 'bicubic' #cv2.INTER_CUBIC
         keys: [image, image]
       - name: PairedRandomCrop
-        output_size: [256, 256]
+        size: [256, 256]
         keys: [image, image]
       - name: PairedRandomHorizontalFlip
         prob: 0.5
         keys: [image, image]
-      - name: Permute
+      - name: Transpose
         keys: [image, image]
       - name: Normalize
         mean: [127.5, 127.5, 127.5]
@@ -62,9 +61,9 @@ dataset:
     transforms:
       - name: Resize
         size: [256, 256]
-        interpolation: 2 #cv2.INTER_CUBIC
+        interpolation: 'bicubic' #cv2.INTER_CUBIC
         keys: [image, image]
-      - name: Permute
+      - name: Transpose
         keys: [image, image]
       - name: Normalize
         mean: [127.5, 127.5, 127.5]

diff --git a/configs/pix2pix_facades.yaml b/configs/pix2pix_facades.yaml
@@ -1,5 +1,4 @@
 epochs: 200
-isTrain: True
 output_dir: output_dir
 lambda_L1: 100
 
@@ -35,15 +34,15 @@ dataset:
     transforms:
       - name: Resize
         size: [286, 286]
-        interpolation: 2 #cv2.INTER_CUBIC
+        interpolation: 'bicubic' #cv2.INTER_CUBIC
         keys: [image, image]
       - name: PairedRandomCrop
-        output_size: [256, 256]
+        size: [256, 256]
         keys: [image, image]
       - name: PairedRandomHorizontalFlip
         prob: 0.5
         keys: [image, image]
-      - name: Permute
+      - name: Transpose
         keys: [image, image]
       - name: Normalize
         mean: [127.5, 127.5, 127.5]
@@ -62,9 +61,9 @@ dataset:
     transforms:
       - name: Resize
         size: [256, 256]
-        interpolation: 2 #cv2.INTER_CUBIC
+        interpolation: 'bicubic' #cv2.INTER_CUBIC
         keys: [image, image]
-      - name: Permute
+      - name: Transpose
         keys: [image, image]
       - name: Normalize
         mean: [127.5, 127.5, 127.5]

diff --git a/ppgan/apps/base_predictor.py b/ppgan/apps/base_predictor.py
@@ -14,6 +14,7 @@
 
 import os
 import cv2
+from PIL import Image
 import paddle
 
 
@@ -61,9 +62,10 @@ def base_forward(self, inputs):
 
         return out
 
-    def is_video(self, input):
+    def is_image(self, input):
         try:
-            cv2.VideoCapture(input)
+            img = Image.open(input)
+            _ = img.size
             return True
         except:
             return False

diff --git a/ppgan/apps/deoldify_predictor.py b/ppgan/apps/deoldify_predictor.py
@@ -128,13 +128,15 @@ def run_video(self, video):
         return frame_pattern_combined, vid_out_path
 
     def run(self, input):
-        if self.is_video(input):
+        if not self.is_image(input):
             return self.run_video(input)
         else:
             pred_img = self.run_image(input)
 
+            out_path = None
             if self.output:
-                base_name = os.path.basename(input)
-                pred_img.save(os.path.join(self.output, base_name + '.png'))
+                base_name = os.path.splitext(os.path.basename(input))[0]
+                out_path = os.path.join(self.output, base_name + '.png')
+                pred_img.save(out_path)
 
-            return pred_img
+            return pred_img, out_path
diff --git a/ppgan/apps/realsr_predictor.py b/ppgan/apps/realsr_predictor.py
@@ -98,13 +98,18 @@ def run_video(self, video):
         return frame_pattern_combined, vid_out_path
 
     def run(self, input):
-        if self.is_video(input):
+        if not os.path.exists(self.output):
+            os.makedirs(self.output)
+
+        if not self.is_image(input):
             return self.run_video(input)
         else:
             pred_img = self.run_image(input)
 
+            out_path = None
             if self.output:
-                base_name = os.path.basename(input)
-                pred_img.save(os.path.join(self.output, base_name + '.png'))
+                base_name = os.path.splitext(os.path.basename(input))[0]
+                out_path = os.path.join(self.output, base_name + '.png')
+                pred_img.save(out_path)
 
-            return pred_img
+            return pred_img, out_path
diff --git a/ppgan/datasets/paired_dataset.py b/ppgan/datasets/paired_dataset.py
@@ -16,7 +16,7 @@ def __init__(self, cfg):
         """Initialize this dataset class.
 
         Args:
-            cfg (dict) -- stores all the experiment flags
+            cfg (dict): configs of datasets.
         """
         BaseDataset.__init__(self, cfg)
         self.dir_AB = os.path.join(cfg.dataroot,
@@ -42,7 +42,7 @@ def __getitem__(self, index):
         """
         # read a image given a random integer index
         AB_path = self.AB_paths[index]
-        AB = cv2.imread(AB_path)
+        AB = cv2.cvtColor(cv2.imread(AB_path), cv2.COLOR_BGR2RGB)
 
         # split AB image into A and B
         h, w = AB.shape[:2]

diff --git a/ppgan/datasets/transforms/__init__.py b/ppgan/datasets/transforms/__init__.py
@@ -1 +1 @@
-from .transforms import RandomCrop, Resize, RandomHorizontalFlip, PairedRandomCrop, PairedRandomHorizontalFlip, Normalize, Permute
+from .transforms import PairedRandomCrop, PairedRandomHorizontalFlip
diff --git a/ppgan/datasets/transforms/builder.py b/ppgan/datasets/transforms/builder.py
@@ -27,6 +27,7 @@ def __call__(self, data):
             try:
                 data = f(data)
             except Exception as e:
+                print(f)
                 stack_info = traceback.format_exc()
                 print("fail to perform transform [{}] with error: "
                       "{} and stack:\n{}".format(f, e, str(stack_info)))

diff --git a/ppgan/datasets/transforms/makeup_transforms.py b/ppgan/datasets/transforms/makeup_transforms.py
@@ -20,7 +20,7 @@ def get_makeup_transform(cfg, pic="image"):
     if pic == "image":
         transform = T.Compose([
             T.Resize(size=cfg.trans_size),
-            T.Permute(to_rgb=False),
+            T.Transpose(),
         ])
     else:
         transform = T.Resize(size=cfg.trans_size,
Original file line number	Diff line number	Diff line change
		@@ -1 +1 @@
		from .transforms import RandomCrop, Resize, RandomHorizontalFlip, PairedRandomCrop, PairedRandomHorizontalFlip, Normalize, Permute
		from .transforms import PairedRandomCrop, PairedRandomHorizontalFlip