import gradio as gr from transformers import Pix2StructForConditionalGeneration, Pix2StructProcessor import requests from PIL import Image import torch torch.hub.download_url_to_file('https://raw.githubusercontent.com/vis-nlp/ChartQA/main/ChartQA%20Dataset/val/png/20294671002019.png', 'chart_example.png') torch.hub.download_url_to_file('https://raw.githubusercontent.com/vis-nlp/ChartQA/main/ChartQA%20Dataset/test/png/multi_col_1081.png', 'chart_example_2.png') torch.hub.download_url_to_file('https://raw.githubusercontent.com/vis-nlp/ChartQA/main/ChartQA%20Dataset/test/png/18143564004789.png', 'chart_example_3.png') torch.hub.download_url_to_file('https://sharkcoder.com/files/article/matplotlib-bar-plot.png', 'chart_example_4.png') model_name = "google/matcha-chartqa" model = Pix2StructForConditionalGeneration.from_pretrained(model_name) processor = Pix2StructProcessor.from_pretrained(model_name) device = torch.device("cuda" if torch.cuda.is_available() else "cpu") model.to(device) def filter_output(output): return output.replace("<0x0A>", "") def chart_qa(image, question): inputs = processor(images=image, text=question, return_tensors="pt").to(device) predictions = model.generate(**inputs, max_new_tokens=512) return filter_output(processor.decode(predictions[0], skip_special_tokens=True)) image = gr.inputs.Image(type="pil", label="Chart") question = gr.inputs.Textbox(label="Question") answer = gr.outputs.Textbox(label="Model Output") examples = [["chart_example.png", "Which country has the second highest death rate?"], ["chart_example_2.png", "Which year has the smallest gap between B2B and B2C sales?"], ["chart_example_3.png", "Which country has the highest CPA received in 2005?"],] # ["chart_example_4.png"]] title = "Interactive demo: Chart QA with MatCha" description = "Gradio Demo for the [matcha](https://arxiv.org/abs/2212.09662) model, fine-tuned on the [ChartQA](https://paperswithcode.com/dataset/chartqa) dataset. To use it, simply upload your image and click 'submit', or click one of the examples to load them." interface = gr.Interface(fn=chart_qa, inputs=[image, question], outputs=answer, examples=examples, title=title, description=description, theme='gradio/soft', enable_queue=True) interface.launch(debug=True)