license: cc-by-nc-4.0
library_name: open_clip
pipeline_tag: image-classification
base_model: HopitAI/moda-fashion-distilled
language:
- en
metrics:
- f1
- accuracy
tags:
- fashion
- attribute-extraction
- product-attributes
- catalog-enrichment
- e-commerce
- computer-vision
- image-classification
- multi-label-classification
- siglip
- moda-ner
- benchmark
- reproducibility
- non-commercial
- color-detection
- fit-prediction
model-index:
- name: MODA_NER(V) Catalog
results:
- task:
type: image-classification
name: Fashion attribute extraction
dataset:
type: moda-general-attribute-suite
name: MODA General Attribute Suite (catalog track)
metrics:
- type: f1
value: 0.8292
name: Field-macro set F1
MODA_NER(V) - Catalog
Tier * - open code, open weights. Licence: CC BY-NC 4.0.
Tier * — open code, open weights. Weights: CC BY-NC 4.0.
Ten field-specific supervised heads on our own frozen encoder (see Provenance below).
Input contract: one clean catalogue product image. Output: category, collar, colour, fabric, fastening, fit, neckline, pattern, pocket, sleeve length.
| Field-macro set F1 (catalog track) | |
|---|---|
| This released checkpoint | 0.8292 |
| Strongest external open baseline | 0.6657 |
Selected best-internal fields: colour 0.7053, fit 0.6708, fabric 0.6693, pocket 0.9491.
Why non-commercial. This track is evaluated against a research-only corpus whose terms do not permit commercial use of models trained on it. We honour those terms, and they bind us as well: these weights are not part of Hopit's hosted product. For commercial deployment we fine-tune on the customer's own catalogue, which raises accuracy on their taxonomy and produces a model with no dependency on research-licensed data.
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) when reporting numbers from this track.
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-catalog --local-dir ./moda-ner-v-catalog
python models/inference.py --route catalog --model-dir ./moda-ner-v-catalog --images photo.jpg