Fashionpedia Dataset
Visual fashion dataset derived from Fashionpedia, adapted to the detection and search of clothing objects (tops, shoes, bags, etc.).
93,600 images (JPEG) with object annotations
Attribution 4.0 International (CC BY 4.0)
Description
The dataset Fashionpedia is a modified version of the original dataset, focused on visual recognition and retrieval of fashion-related objects. It contains over 93,000 street-style images from online catalogs, annotated according to categories such as tops, pants, shoes and accessories (bags, watches, glasses, etc.).
What is this dataset for?
- Training object detection models for fashion
- Building image search engines based on similarity (e.g. CLIP)
- Develop visual clothing recommendation systems
Can it be enriched or improved?
Yes, the dataset can be extended with metadata such as season, style, or materials. It is also possible to link the detected objects to textual descriptions for vision-language learning. Clusters by color or morphology can also improve search or recommendation results.
🔎 In summary
🧠 Recommended for
- AI Startups developing Fashion products
- Fashion search engines
- Specialized CLIP projects
🔧 Compatible tools
- YoloV8
- CLIP
- Roboflow
- Detectron2
- OpenCV
💡 Tip
Combine this dataset with automatically generated text descriptions to create an image-to-text correspondence system.
Frequently Asked Questions
Can this dataset be used to classify clothing styles?
Indirectly yes, by combining the detected objects with rules or a supervised model to classify styles (casual, sport, etc.).
Is this dataset compatible with CLIP models?
Yes, it is specifically designed for visual-textual tasks, such as searching for images by text or vice versa.
Are there annotations for colors or textures?
No, annotations focus on object categories. However, you can extract them via image processing.
