Adapt the resize to cv2 version to keep consistent with that used in training.

app.py

@@ -15,10 +15,17 @@ config = Config()
 device = config.device
 
 
+
+def array_to_pil_image(image, size=(1024, 1024)):
+ image = cv2.resize(image, size, interpolation=cv2.INTER_LINEAR)
+ image = Image.fromarray(image).convert('RGB')
+ return image
+
+
 class ImagePreprocessor():
  def __init__(self, resolution=(1024, 1024)) -> None:
  self.transform_image = transforms.Compose([
- transforms.Resize(resolution),
+ # transforms.Resize(resolution), # 1. keep consistent with the cv2.resize used in training 2. redundant with that in path_to_image()
  transforms.ToTensor(),
  transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]),
  ])
@@ -43,11 +50,12 @@ model.eval()
 # def predict(image_1, image_2):
 # images = [image_1, image_2]
 def predict(image, resolution='1024x1024'):
+ # Image is a RGB numpy array.
+ resolution = [int(int(reso)//32*32) for reso in resolution.strip().split('x')]
  images = [image]
  image_shapes = [image.shape[:2] for image in images]
- images = [Image.fromarray(image) for image in images]
+ images = [array_to_pil_image(image, resolution) for image in images]
 
- resolution = [int(int(reso)//32*32) for reso in resolution.strip().split('x')]
  image_preprocessor = ImagePreprocessor(resolution=resolution)
  images_proc = []
  for image in images: