burman/Image-Caption

1import torch2import gradio as gr 3import re 4from transformers import AutoTokenizer, ViTFeatureExtractor, VisionEncoderDecoderModel5 6device='cpu'7encoder_checkpoint = "nlpconnect/vit-gpt2-image-captioning"8decoder_checkpoint = "nlpconnect/vit-gpt2-image-captioning"9model_checkpoint = "nlpconnect/vit-gpt2-image-captioning"10feature_extractor = ViTFeatureExtractor.from_pretrained(encoder_checkpoint)11tokenizer = AutoTokenizer.from_pretrained(decoder_checkpoint)12model = VisionEncoderDecoderModel.from_pretrained(model_checkpoint).to(device)13 14def predict(image,max_length=64, num_beams=4):15  image = image.convert('RGB')16  image = feature_extractor(image, return_tensors="pt").pixel_values.to(device)17  clean_text = lambda x: x.replace('<|endoftext|>','').split('\n')[0]18  caption_ids = model.generate(image, max_length = max_length)[0]19  caption_text = clean_text(tokenizer.decode(caption_ids))20  return caption_text 21 22def set_example_image(example: list) -> dict:23    return gr.Image.update(value=example[0])24css = '''25h1#title {26  text-align: center;27}28h3#header {29  text-align: center;30}31img#overview {32  max-width: 800px;33  max-height: 600px;34}35img#style-image {36  max-width: 1000px;37  max-height: 600px;38}39'''40demo = gr.Blocks(css=css)41with demo:42  gr.Markdown('''<h1 id="title">Image Caption 🖼️</h1>''')43  gr.Markdown('''Made by : Shreyas Dixit''')44  with gr.Column():45        input = gr.inputs.Image(label="Upload your Image", type = 'pil', optional=True)46        output = gr.outputs.Textbox(type="auto",label="Captions")47  btn = gr.Button("Genrate Caption")48  btn.click(fn=predict, inputs=input, outputs=output)49 50demo.launch()