Update README.md

README.md

@@ -12,24 +12,26 @@ pinned: false
 Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
 
 CLIP Image Recommender (Stanford Online Products) -
+
 This project implements an image-based recommendation system using CLIP embeddings and cosine similarity.
-A user uploads an image, the app converts it into an embedding vector using the same pretrained CLIP model used for the dataset, and then returns the Top-3 most visually similar items from the dataset.
+A user uploads an image, the app converts it into an embedding vector using the same pretrained CLIP model used for the dataset, 
+and then returns the Top-3 most visually similar items from the dataset.
 
 Dataset -
-Source: JamieSJS/stanford-online-products (Hugging Face)
-Modality: images
-Working subset: 3,000 randomly sampled items (for efficiency and reproducibility)
+- Source: JamieSJS/stanford-online-products (Hugging Face)
+- Modality: images
+- Working subset: 3,000 randomly sampled items (for efficiency and reproducibility)
 
 Method - 
-Precompute CLIP image embeddings for a random subset of 3,000 images
-L2-normalize embeddings to support cosine similarity
-At inference time: embed the uploaded image → compute similarity → return Top-3 recommendations
-Results are shown as a gallery + similarity scores
+- Precompute CLIP image embeddings for a random subset of 3,000 images
+- L2-normalize embeddings to support cosine similarity
+- At inference time: embed the uploaded image → compute similarity → return Top-3 recommendations
+- Results are shown as a gallery + similarity scores
 
 Files in the repo -
-app.py – Gradio application
-clip_embeddings_3000.parquet – precomputed normalized embeddings
-sampled_indices_3000.npy – indices of the sampled subset (reproducibility)
+- app.py – Gradio application
+- clip_embeddings_3000.parquet – precomputed normalized embeddings
+- sampled_indices_3000.npy – indices of the sampled subset (reproducibility)
 
 How to use -
 Upload an image → the system returns the 3 most visually similar products.