Confidence score or probability score of Pix2struct VQA model
#2
by
sahumson
- opened
How do we get the confidence score of the predictions for pix2struct model as mentioned below code in pred[0], how do we get the prediction scores?
{
FILENAME = "XXX.pdf"
PAGE_NO = 1
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
model = psg.from_pretrained("google/pix2struct-docvqa-base").to(DEVICE)
processor = psp.from_pretrained("google/pix2struct-docvqa-large")
def convert_pdf_to_image(filename, page_no):
return convert_from_path(filename)[page_no-1]
image = convert_pdf_to_image(FILENAME, PAGE_NO)
inputs = processor(images=[image for _ in range(len(questions))], text=questions, return_tensors="pt").to(DEVICE)
pred=model.generate(**inputs)
print(processor.decode(pred[0],skip_special_tokens=True))
}