Create app.py

app.py

@@ -0,0 +1,72 @@
+import gradio as gr
+from transformers import TrOCRProcessor, VisionEncoderDecoderModel
+import requests
+from PIL import Image
+from craft_text_detector import (
+    read_image,
+    load_craftnet_model,
+    load_refinenet_model,
+    get_prediction,
+    export_detected_regions,
+    export_extra_results,
+    empty_cuda_cache
+)
+
+processor = TrOCRProcessor.from_pretrained("microsoft/trocr-base-handwritten")
+model = VisionEncoderDecoderModel.from_pretrained("microsoft/trocr-base-handwritten")
+craft = Craft(output_dir=None, 
+              crop_type="poly",
+              export_extra=False,
+              link_threshold=0.1,
+              text_threshold=0.3,
+              cuda=torch.cuda.is_available())
+
+
+# load image examples from the IAM database
+urls = ['https://fki.tic.heia-fr.ch/static/img/a01-122-02.jpg', 'https://encrypted-tbn0.gstatic.com/images?q=tbn:ANd9GcSoolxi9yWGAT5SLZShv8vVd0bz47UWRzQC19fDTeE8GmGv_Rn-PCF1pP1rrUx8kOjA4gg&usqp=CAU',
+        'https://encrypted-tbn0.gstatic.com/images?q=tbn:ANd9GcRNYtTuSBpZPV_nkBYPMFwVVD9asZOPgHww4epu9EqWgDmXW--sE2o8og40ZfDGo87j5w&usqp=CAU']
+for idx, url in enumerate(urls):
+  image = Image.open(requests.get(url, stream=True).raw)
+  image.save(f"image_{idx}.png")
+
+def process_image(image):
+    img = np.array(image)
+    prediction_result = craft.detect_text(img)
+    text = []
+    for i,j in enumerate(prediction_result['boxes']): 
+      roi = img[int(prediction_result['boxes'][i][0][1]): int(prediction_result['boxes'][i][2][1]), 
+                int(prediction_result['boxes'][i][0][0]): int(prediction_result['boxes'][i][2][0])]
+      image = Image.fromarray(roi).convert("RGB")
+      pixel_values = processor(image, return_tensors="pt").pixel_values
+      generated_ids = model.generate(pixel_values)
+      generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
+      text.append(generated_text)
+      print('line ' + str(i) + ' has been recoginized')
+
+    generated_text = ('\n').join(text)
+
+
+    # # prepare image
+    # pixel_values = processor(image, return_tensors="pt").pixel_values
+
+    # # generate (no beam search)
+    # generated_ids = model.generate(pixel_values)
+
+    # # decode
+    # generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
+
+    return generated_text
+
+title = "Interactive demo: TrOCR"
+description = "Demo for Microsoft's TrOCR, an encoder-decoder model consisting of an image Transformer encoder and a text Transformer decoder for state-of-the-art optical character recognition (OCR) on single-text line images. This particular model is fine-tuned on IAM, a dataset of annotated handwritten images. To use it, simply upload an image or use the example image below and click 'submit'. Results will show up in a few seconds."
+article = "<p style='text-align: center'><a href='https://arxiv.org/abs/2109.10282'>TrOCR: Transformer-based Optical Character Recognition with Pre-trained Models</a> | <a href='https://github.com/microsoft/unilm/tree/master/trocr'>Github Repo</a></p>"
+examples =[["image_0.png"], ["image_1.png"], ["image_2.png"]]
+
+iface = gr.Interface(fn=process_image, 
+                     inputs=gr.inputs.Image(type="pil"), 
+                     outputs=gr.outputs.Textbox(),
+                     title=title,
+                     description=description,
+                     article=article,
+                     examples=examples)
+iface.launch(debug=True,share=True)