Apppublic

maigurski/maigurski-indoor-outdoor-recsys

sourceHugging Facemitupdated 9mo agoView on Hugging Face

0likes

App README

Indoor/Outdoor Retrieval (Text + Image) — CLIP Embeddings (5,000 sample)

This Space demonstrates visual retrieval over a precomputed embedding index extracted from the dataset IndoorOutdoorNet-20K.

What the app does

—Text → Image retrieval: type a text query and retrieve the most similar images.
—Image → Image retrieval: upload an image and retrieve similar items from the dataset sample.
—Uses CLIP to embed the user input (text or image), then computes cosine similarity against the stored embeddings.
—Returns Top-K results with similarity scores.

Dataset

—Dataset: IndoorOutdoorNet-20K
—Working sample: 5,000 images (kept small for stable runtime)

Artifacts (one file only)

This Space uses a single artifact file:

—artifacts/embeddings.parquet

The parquet contains:

—orig_idx (original dataset index)
—label (indoor/outdoor)
—the embedding vector stored either as:
—a single embedding column, OR
—wide columns emb_0 ... emb_{D-1}

The app supports both formats.

A) Put the video in the README

Replace YOUR_VIDEO_ID:

html

<iframe width="560" height="315"
    src=https://youtu.be/1Jo-H3WvO7c
    title="YouTube video player"
    frameborder="0"
    allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture"
    allowfullscreen>
</iframe>