moda-ner-v-fullbody

MODA_NER(V) - Full-body

Fix, 2026-09-19. If you ran this model before this date, re-run it. The repository's models/inference.py ran the full-body route with the wrong image preprocessing: OpenAI-CLIP normalisation and a centre crop, instead of the mean/std 0.5 squash this checkpoint was trained with. Its outputs did not match the published benchmark predictions, matching on none of 64 held-out images. The weights in this repository are unchanged and correct. Pull the latest Moda_ner code and the Usage command below produces the published predictions. The benchmark numbers on this card were produced correctly.

Tier * - open code, open weights. Licence: CC BY-NC 4.0.

Tier * — open code, open weights. Weights: CC BY-NC 4.0.

Eighteen class-balanced heads with an explicit N/A class, on our own frozen encoder (see Provenance below).

Input contract: one full-body fashion photograph. Output: upper, lower and outer fabric; upper, lower and outer pattern; neckline; sleeve length; lower-garment length; cardigan; navel coverage; hat; glasses; neckwear; waist accessories; wrist accessories; ring; socks or leggings. Every field can return N/A, and N/A is scored as a real class rather than treated as a negative.

These are the weights that produced the published numbers above (run ...linear-probe-v7-20260810, recorded in training_commitment.json). Evaluated on a fresh product-group-disjoint split: 5,000 images, 1,751 groups, zero overlap with any prior experiment. Per-field range is wide (0.3287 to 0.9187) — read the per-field table before relying on a single field.

Why non-commercial. Same as the catalog route: the evaluation corpus is research-only and its terms extend to derived data. These weights are not in Hopit's hosted product. Commercial deployments are fine-tuned on customer data instead.

Provenance

The encoder these heads run on is ours: HopitAI/moda-fashion-distilled, MIT, already public. Nothing from another vendor is loaded at inference time. (Recorded in the programme documentation; we re-confirm it against this track's frozen artifacts before release.)

That is worth stating plainly, because the comparator on this track is a FashionSigLIP-based system and it would be easy to assume this model is that system with heads attached. It is not. FashionSigLIP appears in two other roles:

• As the distillation teacher. An earlier ladder of checkpoints put conditional heads on frozen Marqo-FashionSigLIP. We distilled that system into our own encoder; the teacher is used during training and is not needed to serve.
• As the baseline we measure against. The comparator figure quoted above is that same FashionSigLIP-based system.

Lineage, stated once rather than implied: moda-fashion-distilled is itself a distilled student built on ViT-B/16-SigLIP, from a teacher ensemble that included our own DeepFashion2 fine-tune. Marqo-FashionSigLIP is Apache-2.0. The DeepFashion2 corpus is research-only, so we do not describe this pipeline as provenance-clean end to end.

Credit for this model. CC BY-NC requires attribution. Cite the MODA General Attribute Suite (CITATION.cff).

Links

• Benchmark tables and protocol: https://hopit-ai.github.io/Moda_ner/
• Code, scorers and prediction files: https://github.com/hopit-ai/Moda_ner
• All Hopit AI benchmarks: https://hopit-ai.github.io/

Usage

The heads are not a transformers architecture, so load them through the suite repository rather than AutoModel:

git clone https://github.com/hopit-ai/Moda_ner && cd Moda_ner
pip install -r requirements-inference.txt
huggingface-cli download HopitAI/moda-ner-v-fullbody --local-dir ./moda-ner-v-fullbody
python models/inference.py --route fullbody --model-dir ./moda-ner-v-fullbody --images photo.jpg