Upload app.py

app.py

@@ -0,0 +1,109 @@
+# app.py
+
+# 🛠️ Setup
+# pip install -q gradio torch ftfy regex tqdm git+https://github.com/openai/CLIP.git matplotlib
+
+# 📦 Imports
+import gradio as gr
+import torch
+import clip
+from PIL import Image
+import numpy as np
+from typing import List, Tuple, Union
+
+# 🚀 Load CLIP Model
+device: str = "cuda" if torch.cuda.is_available() else "cpu"
+model, preprocess = clip.load("ViT-B/32", device=device)
+
+
+# def print_installed_packages():
+#     installed_packages = pip.get_installed_distributions()
+#     for package in installed_packages:
+#         print(f"{package.project_name}=={package.version}")
+
+def predict(image: Image.Image, label_text: str) -> List[List[Union[str, float]]]:
+    """
+    Perform zero-shot classification using the CLIP model.
+
+    Args:
+        image (PIL.Image.Image): Input image.
+        label_text (str): Comma-separated labels to classify against.
+
+    Returns:
+        List[List[Union[str, float]]]: A list of results with label, probability, and confidence bar HTML.
+    """
+    labels: List[str] = [label.strip() for label in label_text.split(",") if label.strip()]
+    if not image or not labels:
+        return []
+
+    # Preprocess inputs
+    image_input: torch.Tensor = preprocess(image).unsqueeze(0).to(device)
+    text_inputs: torch.Tensor = clip.tokenize(labels).to(device)
+
+    # Run model
+    with torch.no_grad():
+        image_features: torch.Tensor = model.encode_image(image_input)
+        text_features: torch.Tensor = model.encode_text(text_inputs)
+        logits_per_image, _ = model(image_input, text_inputs)
+        probs: np.ndarray = logits_per_image.softmax(dim=-1).cpu().numpy()[0]
+
+    # Create table with bar visualization
+    results: List[List[Union[str, float]]] = []
+    for label, prob in zip(labels, probs):
+        bar_html: str = (
+            f'<div style="background-color:#4caf50;width:{prob * 100:.1f}%;height:20px;"></div>'
+        )
+        results.append([label, f"{prob * 100:.2f}%", bar_html])
+
+    return results
+
+
+# 🎨 Gradio Interface
+with gr.Blocks() as demo:
+    gr.Markdown("## CLIP Zero-Shot Classifier")
+
+    with gr.Row():
+        image = gr.Image(type="pil", label="Upload Image")
+        label_text = gr.Textbox(
+            lines=2,
+            label="Enter comma-separated labels",
+            placeholder="e.g., a cat, a dog, a diagram"
+        )
+
+    # Image Examples
+    with gr.Row():
+        gr.Examples(
+            examples=[
+                ["images/boy.jpg"],
+                ["images/dog.jpg"],
+                ["images/boy_dog.jpg"]
+            ],
+            inputs=[image],
+            label="🖼️ Click to select example image"
+        )
+
+        # Label Text Examples
+        gr.Examples(
+            examples=[
+                ["boy, girl, dog, cat"],
+                ["a boy with a dog, a boy with a cat, a girl with a dog, a girl with a cat"],
+                ["a cat, a dog, a diagram"]
+            ],
+            inputs=[label_text],
+            label="📝 Click to autofill example labels"
+        )
+
+    submit = gr.Button("Classify")
+
+    output = gr.Dataframe(
+        headers=["Label", "Probability", "Confidence Bar"],
+        datatype=["str", "str", "html"],
+        row_count=5,
+        interactive=False
+    )
+
+    submit.click(fn=predict, inputs=[image, label_text], outputs=output)
+
+if __name__ == "__main__":
+    # print_installed_packages()
+    demo.launch(share=True)