Create app.py

First version of MatchMe app — upload an image and get fashion recommendations.

app.py

@@ -0,0 +1,52 @@
+import gradio as gr
+import torch
+import pickle
+import numpy as np
+from PIL import Image
+from torchvision import transforms
+from huggingface_hub import hf_hub_download
+from sklearn.metrics.pairwise import cosine_similarity
+
+# Step 1: Load the precomputed fashion embeddings
+file_path = hf_hub_download(repo_id="Elevi7/MatchMe", filename="fashion_embeddings.pkl", repo_type="dataset")
+with open(file_path, "rb") as f:
+    embedding_store = pickle.load(f)
+
+# Step 2: Load the same model used to create the embeddings
+# (update this if you used something different, e.g. CLIP, ResNet, etc.)
+import timm
+model = timm.create_model("resnet18", pretrained=True)
+model.eval()
+
+transform = transforms.Compose([
+    transforms.Resize((224, 224)),
+    transforms.ToTensor()
+])
+
+def extract_embedding(image):
+    image = transform(image).unsqueeze(0)
+    with torch.no_grad():
+        embedding = model(image).squeeze().numpy()
+    return embedding
+
+def recommend(image):
+    query_embedding = extract_embedding(image)
+
+    # Get embeddings and paths
+    all_embeddings = np.array([item["embedding"] for item in embedding_store])
+    paths = [item["image_path"] for item in embedding_store]
+
+    similarities = cosine_similarity([query_embedding], all_embeddings)[0]
+    top_indices = np.argsort(similarities)[-3:][::-1]  # Top 3 matches
+
+    return [paths[i] for i in top_indices]
+
+demo = gr.Interface(
+    fn=recommend,
+    inputs=gr.Image(type="pil"),
+    outputs=[gr.Image(type="filepath", label=f"Match {i+1}") for i in range(3)],
+    title="MatchMe: Fashion Recommender",
+    description="Upload a fashion image and get 3 visually similar items."
+)
+
+demo.launch()