Update app.py

app.py

@@ -126,9 +126,18 @@ def predict(model_name, input_img):
     else:
         return input_img, {"text": text, "score": score}
 
+def get_dynamic_imgsz(image, target_height=640, stride=32):
+    orig_width, orig_height = image.size
+    aspect_ratio = orig_width / orig_height
+    target_width = int(target_height * aspect_ratio)
+    # Round width to next multiple of stride (usually 32 for YOLO)
+    target_width = ((target_width + stride - 1) // stride) * stride
+    return [target_height, target_width]  # YOLO expects [H, W]
+
 def process_image(image):
     # Perform inference on an image, select textline only
-    results = model(image, imgsz=[640, 512], classes=0)
+    imgsz = get_dynamic_imgsz(image)  # dynamically compute correct size
+    results = model(image, classes=0, imgsz=imgsz)
 
     img_cv2 = cv2.cvtColor(np.array(image), cv2.COLOR_RGB2BGR)
     masks = results[0].masks