Spaces:
Paused
Paused
Update app.py
Browse files
app.py
CHANGED
@@ -1,7 +1,49 @@
|
|
1 |
-
torch
|
2 |
-
transformers
|
3 |
-
|
4 |
-
|
5 |
-
|
6 |
-
|
7 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
from transformers import VisionEncoderDecoderModel, DonutProcessor
|
3 |
+
from PIL import Image
|
4 |
+
from pdf2image import convert_from_bytes
|
5 |
+
import gradio as gr
|
6 |
+
|
7 |
+
# Configuraci贸n del modelo Donut
|
8 |
+
MODEL_ID = "mychen76/invoice-and-receipts_donut_v1" # Cambia si usas otro modelo
|
9 |
+
print("Cargando modelo Donut...")
|
10 |
+
model = VisionEncoderDecoderModel.from_pretrained(MODEL_ID)
|
11 |
+
processor = DonutProcessor.from_pretrained(MODEL_ID)
|
12 |
+
model.eval()
|
13 |
+
|
14 |
+
# Funci贸n para procesar documentos
|
15 |
+
def process_document(file):
|
16 |
+
# Si es un PDF, convi茅rtelo a im谩genes
|
17 |
+
if file.name.endswith(".pdf"):
|
18 |
+
pdf_data = file.read()
|
19 |
+
images = convert_from_bytes(pdf_data, dpi=300)
|
20 |
+
else:
|
21 |
+
# Si es imagen, simplemente c谩rgala
|
22 |
+
images = [Image.open(file).convert("RGB")]
|
23 |
+
|
24 |
+
results = []
|
25 |
+
for img in images:
|
26 |
+
# Preprocesar la imagen
|
27 |
+
inputs = processor(img, return_tensors="pt", max_patches=1024)
|
28 |
+
# Generar predicciones
|
29 |
+
with torch.no_grad():
|
30 |
+
outputs = model.generate(**inputs)
|
31 |
+
# Decodificar resultado
|
32 |
+
result = processor.batch_decode(outputs, skip_special_tokens=True)[0]
|
33 |
+
results.append(result)
|
34 |
+
|
35 |
+
# Combina los resultados si son m煤ltiples p谩ginas
|
36 |
+
return results
|
37 |
+
|
38 |
+
# Interfaz Gradio
|
39 |
+
iface = gr.Interface(
|
40 |
+
fn=process_document,
|
41 |
+
inputs=gr.File(label="Sube tu factura o recibo (PDF o imagen)", type="file"),
|
42 |
+
outputs="json",
|
43 |
+
title="Donut OCR - Extracci贸n de datos de facturas",
|
44 |
+
description="Sube un PDF o imagen y extrae informaci贸n estructurada (n煤mero de factura, fecha, monto, etc.) utilizando Donut OCR."
|
45 |
+
)
|
46 |
+
|
47 |
+
# Iniciar la aplicaci贸n
|
48 |
+
if __name__ == "__main__":
|
49 |
+
iface.launch()
|