metadata

dataset_info:
  features:
    - name: image
      dtype: image
    - name: category1
      dtype: string
    - name: category2
      dtype: string
    - name: category3
      dtype: string
    - name: text
      dtype: string
    - name: item_ID
      dtype: string
  splits:
    - name: data
      num_bytes: 4235530431.72
      num_examples: 201624
  download_size: 3466991670
  dataset_size: 4235530431.72
configs:
  - config_name: default
    data_files:
      - split: data
        path: data/data-*
license: apache-2.0

Disclaimer: We do not own this dataset. Fashion200K dataset is a public dataset which can be accessed through its Github page.

This dataset was used to evaluate Marqo-FashionCLIP and Marqo-FashionSigLIP - see details below.

Marqo-FashionSigLIP Model Card

Marqo-FashionSigLIP leverages Generalised Contrastive Learning (GCL) which allows the model to be trained on not just text descriptions but also categories, style, colors, materials, keywords and fine-details to provide highly relevant search results on fashion products. The model was fine-tuned from ViT-B-16-SigLIP (webli).

Github Page: Marqo-FashionCLIP

Blog: Marqo Blog

Usage

The model can be seamlessly used with OpenCLIP by

import open_clip
model, preprocess_train, preprocess_val = open_clip.create_model_and_transforms('hf-hub:Marqo/marqo-fashionSigLIP')
tokenizer = open_clip.get_tokenizer('hf-hub:Marqo/marqo-fashionSigLIP')
import torch
from PIL import Image
image = preprocess_val(Image.open("docs/fashion-hippo.png")).unsqueeze(0)
text = tokenizer(["a hat", "a t-shirt", "shoes"])
with torch.no_grad(), torch.cuda.amp.autocast():
    image_features = model.encode_image(image)
    text_features = model.encode_text(text)
    image_features /= image_features.norm(dim=-1, keepdim=True)
    text_features /= text_features.norm(dim=-1, keepdim=True)
    text_probs = (100.0 * image_features @ text_features.T).softmax(dim=-1)
print("Label probs:", text_probs)

Benchmark Results

Average evaluation results on 6 public multimodal fashion datasets (Atlas, DeepFashion (In-shop), DeepFashion (Multimodal), Fashion200k, KAGL, and Polyvore) are reported below:

Text-To-Image (Averaged across 6 datasets)

Model	AvgRecall	Recall@1	Recall@10	MRR
Marqo-FashionSigLIP	0.231	0.121	0.340	0.239
FashionCLIP2.0	0.163	0.077	0.249	0.165
OpenFashionCLIP	0.132	0.060	0.204	0.135
ViT-B-16-laion2b_s34b_b88k	0.174	0.088	0.261	0.180
ViT-B-16-SigLIP-webli	0.212	0.111	0.314	0.214

Category-To-Product (Averaged across 5 datasets)

Model	AvgP	P@1	P@10	MRR
Marqo-FashionSigLIP	0.737	0.758	0.716	0.812
FashionCLIP2.0	0.684	0.681	0.686	0.741
OpenFashionCLIP	0.646	0.653	0.639	0.720
ViT-B-16-laion2b_s34b_b88k	0.662	0.673	0.652	0.743
ViT-B-16-SigLIP-webli	0.688	0.690	0.685	0.751

Sub-Category-To-Product (Averaged across 4 datasets)

Model	AvgP	P@1	P@10	MRR
Marqo-FashionSigLIP	0.725	0.767	0.683	0.811
FashionCLIP2.0	0.657	0.676	0.638	0.733
OpenFashionCLIP	0.598	0.619	0.578	0.689
ViT-B-16-laion2b_s34b_b88k	0.638	0.651	0.624	0.712
ViT-B-16-SigLIP-webli	0.643	0.643	0.643	0.726

When using the datset, cite the original work.

@inproceedings{han2017automatic,
  title = {Automatic Spatially-aware Fashion Concept Discovery},
  author = {Han, Xintong and Wu, Zuxuan and Huang, Phoenix X. and Zhang, Xiao and Zhu, Menglong and Li, Yuan and Zhao, Yang  and Davis, Larry S.},
  booktitle = {ICCV},
  year  = {2017},
}