Spaces:
Sleeping
Sleeping
import gradio as gr | |
from transformers import TrOCRProcessor, VisionEncoderDecoderModel | |
import requests | |
from PIL import Image | |
import numpy as np | |
import cv2 | |
processor = TrOCRProcessor.from_pretrained("microsoft/trocr-base-handwritten") | |
model = VisionEncoderDecoderModel.from_pretrained("aico/TrOCR-MNIST") | |
def process_image(image): | |
#print(np.shape(image)) | |
#print(image) | |
#rint(image.astype('uint8')) | |
#cv2.imwrite("image.png",image.astype('uint8'),(28, 28)) | |
img = Image.fromarray(image.astype('uint8')).convert("RGB") | |
#img = Image.open("image.png").convert("RGB") | |
print(img) | |
# prepare image | |
pixel_values = processor(img, 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: Single Digits MNIST" | |
description = "Aico - University Utrecht" | |
iface = gr.Interface(fn=process_image, | |
inputs="sketchpad", | |
outputs="label", | |
title = title, | |
description = description) | |
iface.launch(debug=True) |