Update README.md
Browse files
README.md
CHANGED
|
@@ -36,4 +36,66 @@ configs:
|
|
| 36 |
path: data/data-*
|
| 37 |
license: mit
|
| 38 |
---
|
| 39 |
-
**Disclaimer**: We do not own this dataset. KAGL dataset is a public dataset which can be accessed through [Kaggle](https://www.kaggle.com/datasets/paramaggarwal/fashion-product-images-dataset).
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 36 |
path: data/data-*
|
| 37 |
license: mit
|
| 38 |
---
|
| 39 |
+
**Disclaimer**: We do not own this dataset. KAGL dataset is a public dataset which can be accessed through [Kaggle](https://www.kaggle.com/datasets/paramaggarwal/fashion-product-images-dataset).
|
| 40 |
+
|
| 41 |
+
This dataset was used to evaluate Marqo-FashionCLIP and Marqo-FashionSigLIP - see details below.
|
| 42 |
+
|
| 43 |
+
# Marqo-FashionSigLIP Model Card
|
| 44 |
+
Marqo-FashionSigLIP leverages Generalised Contrastive Learning ([GCL](https://www.marqo.ai/blog/generalized-contrastive-learning-for-multi-modal-retrieval-and-ranking)) 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.
|
| 45 |
+
The model was fine-tuned from ViT-B-16-SigLIP (webli).
|
| 46 |
+
|
| 47 |
+
**Github Page**: [Marqo-FashionCLIP](https://github.com/marqo-ai/marqo-FashionCLIP)
|
| 48 |
+
|
| 49 |
+
**Blog**: [Marqo Blog](https://www.marqo.ai/blog/search-model-for-fashion)
|
| 50 |
+
|
| 51 |
+
|
| 52 |
+
## Usage
|
| 53 |
+
The model can be seamlessly used with [OpenCLIP](https://github.com/mlfoundations/open_clip) by
|
| 54 |
+
|
| 55 |
+
```python
|
| 56 |
+
import open_clip
|
| 57 |
+
model, preprocess_train, preprocess_val = open_clip.create_model_and_transforms('hf-hub:Marqo/marqo-fashionSigLIP')
|
| 58 |
+
tokenizer = open_clip.get_tokenizer('hf-hub:Marqo/marqo-fashionSigLIP')
|
| 59 |
+
import torch
|
| 60 |
+
from PIL import Image
|
| 61 |
+
image = preprocess_val(Image.open("docs/fashion-hippo.png")).unsqueeze(0)
|
| 62 |
+
text = tokenizer(["a hat", "a t-shirt", "shoes"])
|
| 63 |
+
with torch.no_grad(), torch.cuda.amp.autocast():
|
| 64 |
+
image_features = model.encode_image(image)
|
| 65 |
+
text_features = model.encode_text(text)
|
| 66 |
+
image_features /= image_features.norm(dim=-1, keepdim=True)
|
| 67 |
+
text_features /= text_features.norm(dim=-1, keepdim=True)
|
| 68 |
+
text_probs = (100.0 * image_features @ text_features.T).softmax(dim=-1)
|
| 69 |
+
print("Label probs:", text_probs)
|
| 70 |
+
```
|
| 71 |
+
|
| 72 |
+
## Benchmark Results
|
| 73 |
+
Average evaluation results on 6 public multimodal fashion datasets ([Atlas](https://huggingface.co/datasets/Marqo/atlas), [DeepFashion (In-shop)](https://huggingface.co/datasets/Marqo/deepfashion-inshop), [DeepFashion (Multimodal)](https://huggingface.co/datasets/Marqo/deepfashion-multimodal), [Fashion200k](https://huggingface.co/datasets/Marqo/fashion200k), [KAGL](https://huggingface.co/datasets/Marqo/KAGL), and [Polyvore](https://huggingface.co/datasets/Marqo/polyvore)) are reported below:
|
| 74 |
+
|
| 75 |
+
**Text-To-Image (Averaged across 6 datasets)**
|
| 76 |
+
| Model | AvgRecall | Recall@1 | Recall@10 | MRR |
|
| 77 |
+
|----------------------------|-------------|------------|-------------|-----------|
|
| 78 |
+
| Marqo-FashionSigLIP | **0.231** | **0.121** | **0.340** | **0.239** |
|
| 79 |
+
| FashionCLIP2.0 | 0.163 | 0.077 | 0.249 | 0.165 |
|
| 80 |
+
| OpenFashionCLIP | 0.132 | 0.060 | 0.204 | 0.135 |
|
| 81 |
+
| ViT-B-16-laion2b_s34b_b88k | 0.174 | 0.088 | 0.261 | 0.180 |
|
| 82 |
+
| ViT-B-16-SigLIP-webli | 0.212 | 0.111 | 0.314 | 0.214 |
|
| 83 |
+
|
| 84 |
+
**Category-To-Product (Averaged across 5 datasets)**
|
| 85 |
+
| Model | AvgP | P@1 | P@10 | MRR |
|
| 86 |
+
|----------------------------|-----------|-----------|-----------|-----------|
|
| 87 |
+
| Marqo-FashionSigLIP | **0.737** | **0.758** | **0.716** | **0.812** |
|
| 88 |
+
| FashionCLIP2.0 | 0.684 | 0.681 | 0.686 | 0.741 |
|
| 89 |
+
| OpenFashionCLIP | 0.646 | 0.653 | 0.639 | 0.720 |
|
| 90 |
+
| ViT-B-16-laion2b_s34b_b88k | 0.662 | 0.673 | 0.652 | 0.743 |
|
| 91 |
+
| ViT-B-16-SigLIP-webli | 0.688 | 0.690 | 0.685 | 0.751 |
|
| 92 |
+
|
| 93 |
+
**Sub-Category-To-Product (Averaged across 4 datasets)**
|
| 94 |
+
| Model | AvgP | P@1 | P@10 | MRR |
|
| 95 |
+
|----------------------------|-----------|-----------|-----------|-----------|
|
| 96 |
+
| Marqo-FashionSigLIP | **0.725** | **0.767** | **0.683** | **0.811** |
|
| 97 |
+
| FashionCLIP2.0 | 0.657 | 0.676 | 0.638 | 0.733 |
|
| 98 |
+
| OpenFashionCLIP | 0.598 | 0.619 | 0.578 | 0.689 |
|
| 99 |
+
| ViT-B-16-laion2b_s34b_b88k | 0.638 | 0.651 | 0.624 | 0.712 |
|
| 100 |
+
| ViT-B-16-SigLIP-webli | 0.643 | 0.643 | 0.643 | 0.726 |
|
| 101 |
+
|