| from transformers import pipeline | |
| import gradio as gr | |
| clip_models = [ | |
| "zer0int/CLIP-GmP-ViT-L-14", | |
| "John6666/zer0int_CLIP-GmP-ViT-L-14", | |
| "openai/clip-vit-large-patch14", | |
| "laion/CLIP-ViT-bigG-14-laion2B-39B-b160k", | |
| ] | |
| clip_checkpoint = clip_models[0] | |
| clip_detector = pipeline(model=clip_checkpoint, task="zero-shot-image-classification") | |
| def postprocess(output): | |
| return {out["label"]: float(out["score"]) for out in output} | |
| def infer(image, candidate_labels): | |
| candidate_labels = [label.lstrip(" ") for label in candidate_labels.split(",")] | |
| clip_out = clip_detector(image, candidate_labels=candidate_labels) | |
| return postprocess(clip_out) | |
| def load_clip_model(modelname): | |
| global clip_detector | |
| try: | |
| clip_detector = pipeline(model=modelname, task="zero-shot-image-classification") | |
| except Exception as e: | |
| raise gr.Error(f"Model load error: {modelname} {e}") | |
| return modelname | |
| with gr.Blocks() as demo: | |
| gr.Markdown("# Test CLIP") | |
| with gr.Row(): | |
| with gr.Column(): | |
| image_input = gr.Image(type="pil") | |
| text_input = gr.Textbox(label="Input a list of labels") | |
| model_input = gr.Dropdown(label="CLIP model", choices=clip_models, value=clip_models[0], allow_custom_value=True, interactive=True) | |
| run_button = gr.Button("Run", visible=True) | |
| with gr.Column(): | |
| clip_output = gr.Label(label = "CLIP Output", num_top_classes=3) | |
| examples = [["./baklava.jpg", "baklava, souffle, tiramisu"], ["./cheetah.jpg", "cat, dog"], ["./cat.png", "cat, dog"]] | |
| gr.Examples( | |
| examples = examples, | |
| inputs=[image_input, text_input], | |
| outputs=[clip_output], | |
| fn=infer, | |
| cache_examples=True | |
| ) | |
| run_button.click(fn=infer, | |
| inputs=[image_input, text_input], | |
| outputs=[clip_output]) | |
| model_input.change(load_clip_model, [model_input], [model_input]) | |
| demo.launch() |