Upload compare_clip_siglip/app.py with huggingface_hub

compare_clip_siglip/app.py

@@ -0,0 +1,46 @@
+import torch
+from transformers import pipeline, SiglipModel, AutoProcessor
+import numpy as np
+import gradio as gr
+
+
+clip_checkpoint = "laion/CLIP-ViT-H-14-laion2B-s32B-b79K"
+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)
+
+with gr.Blocks() as demo:
+  gr.Markdown("# Compare CLIP and SigLIP")
+  gr.Markdown("Compare the performance of CLIP and SigLIP on zero-shot classification in this Space 👇")
+  with gr.Row():
+    with gr.Column():
+        image_input = gr.Image(type="pil")
+        text_input = gr.Textbox(label="Input a list of labels")
+        run_button = gr.Button("Run", visible=True)
+
+    with gr.Column():
+      clip_output = gr.Label(label = "CLIP Output", num_top_classes=15)
+      
+  examples = [["./baklava.jpg", "baklava, souffle, tiramisu"]]
+  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, 
+                             ])
+
+demo.launch()