MODA Pro Lite+

The strongest open system at ≤250M parameters on catalogue and title search.

MODA Pro Lite+ is MODA Pro Lite — a 213M fashion retrieval encoder — served with a calibrated multi-view recipe. This repository holds the recipe, as runnable code; the weights are pulled from moda-pro-lite at load time.

Zero added parameters. One stored vector per item. The uplift is paid once at index time and costs nothing at query time.

Results

MAP@10, full corpus, all ground-truth queries, one evaluator (pytrec_eval map_cut.10). MODA is FashionSigLIP with its own serving recipe, shown for reference.

Pro Lite+ leads the ≤250M class on KAGL, Polyvore and Atlas — +10.9% over MODA on KAGL, +8.7% on Polyvore, both significant under a paired bootstrap (10,000 resamples).

The recipe is worth +2.5% to +12.8% over the bare encoder on every benchmark, and costs nothing at query time: the views are fused into a single vector before indexing.

Where this model is weak, stated plainly. Pro Lite is tuned for short catalogue titles. On long natural-language descriptions it trails FashionSigLIP substantially — DeepFashion In-Shop queries average 75 words, and Pro Lite+ scores 0.1026 there against MODA's 0.1642. If your queries are descriptions rather than titles, use MODA Duo, which routes per query.

Serving cost

stored vectors per item : 1
ANN queries per search  : 1
image forwards at index : 3x   offline, paid once
text forwards per query : 2x   negligible beside the ANN probe

The recipe is a rule for what you encode, not a model change. Views are combined into one unit vector before indexing, so nearest-neighbour search costs exactly what the bare encoder costs — same index, same probe, no extra routes and no re-ranking.

Use

pip install open_clip_torch pillow numpy hnswlib
python serving_ann.py --demo

from serving_ann import load, encode_images, encode_queries, build_index, search

enc   = load()                          # open_clip, this repo's weights
docs  = encode_images(catalogue, enc)   # (n, 768) float32, one vector per item
index = build_index(docs)               # hnswlib, cosine via inner product
qry   = encode_queries(["black leather ankle boots"], enc)
ids, scores = search(index, qry, k=10)

Bare encoder, if you would rather not use the recipe:

import open_clip, torch

model, _, preprocess = open_clip.create_model_and_transforms("hf-hub:HopitAI/moda-pro-lite")
tokenizer = open_clip.get_tokenizer("hf-hub:HopitAI/moda-pro-lite")
model.eval()

with torch.no_grad():
    image = torch.nn.functional.normalize(model.encode_image(preprocess(img).unsqueeze(0)), dim=-1)
    text  = torch.nn.functional.normalize(model.encode_text(tokenizer(["black leather ankle boots"])), dim=-1)
    score = (text @ image.T).item()

768-d embeddings, cosine similarity, one vector per item. Index them in any vector database.

The recipe

document = normalize(official + 0.25 * square_pad + 0.25 * foreground_pad)
query    = normalize(raw      + 0.25 * "a photo of {query}")

serving_ann.py implements it. Zero added parameters, one stored vector.

Evaluation

All figures are full corpus, all ground-truth queries, MAP@10 under one evaluator (pytrec_eval map_cut.10), float32. Per-query results and confidence intervals are in the repository.

Related

• MODA Pro Lite — the bare encoder these weights come from.
• MODA Duo — routes each query to Pro Lite+ or MODA by its shape; beats both on a mixed workload.
• MODA — FashionSigLIP with a serving recipe. Stronger on long descriptions.
• MODA-SigLIP-Distilled — image-to-image retrieval.