Add app

app.py

@@ -0,0 +1,41 @@
+import numpy as np
+import gradio as gr
+from transformers import AutoFeatureExtractor, AutoTokenizer, VisionEncoderDecoderModel
+import re
+import jaconv
+
+#load model
+model_path = "kha-white/manga-ocr-base"
+feature_extractor = AutoFeatureExtractor.from_pretrained(model_path)
+tokenizer = AutoTokenizer.from_pretrained(model_path)
+model = VisionEncoderDecoderModel.from_pretrained(model_path)
+
+
+def post_process(text):
+    text = ''.join(text.split())
+    text = text.replace('…', '...')
+    text = re.sub('[・.]{2,}', lambda x: (x.end() - x.start()) * '.', text)
+    text = jaconv.h2z(text, ascii=True, digit=True)
+    return text
+
+def infer(image):
+    image = image.convert('L').convert('RGB')
+    pixel_values = feature_extractor(image, return_tensors="pt").pixel_values
+    ouput = model.generate(pixel_values)[0]
+    text = tokenizer.decode(ouput, skip_special_tokens=True)
+    text = post_process(text)
+    return text
+
+
+iface = gr.Interface(
+    fn=infer,
+    inputs=[gr.inputs.Image(label="Input", type="pil")],
+    outputs="text",
+    layout="horizontal",
+    theme="huggingface",
+    title="Optical character recognition for Japanese text",
+    description="A simple interface for OCR from Japanese manga",
+    article= "Author: <a href=\"https://huggingface.co/vumichien\">Vu Minh Chien</a>. ",
+    allow_flagging='never',
+)
+iface.launch(enable_queue=True, share=True, debug=True)

requirements.txt

@@ -0,0 +1,3 @@
+fugashi
+unidic_lite
+jaconv