anh trong suot

app.py

@@ -1,5 +1,6 @@
 import gradio as gr
 from transformers import TrOCRProcessor, VisionEncoderDecoderModel
+from PIL import Image
 
 name = "chanelcolgate/trocr-base-printed_captcha_ocr"
 model = VisionEncoderDecoderModel.from_pretrained(name)
@@ -7,13 +8,16 @@ processor = TrOCRProcessor.from_pretrained("microsoft/trocr-base-printed")
 
 
 def process_image(image):
-    # prepare image
+    # Chuyển ảnh sang RGB và thêm nền trắng nếu ảnh có alpha
+    image = image.convert("RGBA")
+    background = Image.new("RGBA", image.size, (255, 255, 255, 255))  # trắng hoàn toàn
+    image = Image.alpha_composite(background, image).convert("RGB")   # merge và chuyển sang RGB
+
+    # Chuẩn bị ảnh cho model
     pixel_values = processor(image, return_tensors="pt").pixel_values
 
-    # generate (no beam search)
+    # Sinh kết quả
     generated_ids = model.generate(pixel_values)
-
-    # decode
     generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
     return generated_text
 
@@ -21,9 +25,13 @@ def process_image(image):
 title = "Interactive demo: Captcha OCR"
 description = "Demo tracuumasothue captcha"
 
-interface = gr.Interface(fn=process_image,
-                         inputs="image",
-                         examples=[f"examples/captcha-{i}.png" for i in range(10)],
-                         outputs="text",
-                         title=title,
-                         description=description).launch()
+interface = gr.Interface(
+    fn=process_image,
+    inputs="image",
+    examples=[f"examples/captcha-{i}.png" for i in range(10)],
+    outputs="text",
+    title=title,
+    description=description
+)
+
+interface.launch()