Detect Fashion Items & Clothing Details
Detects and localizes 46 clothing items and fashion components in images. Fine-tuned version of YOLOS specifically for fashion domain. Perfect for e-commerce product analysis, fashion inventory, and outfit composition understanding.
What it does
YOLOS Fashion detects clothing items and fashion-related components in images. Unlike general-purpose object detectors that struggle with fine-grained fashion details, this model is specialized to identify 46 different fashion elements. It finds both garments (what you're wearing) and details (how they're designed). Point it at any fashion photo, and it returns precise locations and classifications of all detected clothing items and components with confidence scores.
Problem it solves
- E-commerce product analysis – Automatically catalog and analyze fashion product photos
- Clothing detection – Identify garment types (shirts, pants, dresses, jackets, etc.)
- Fashion accessory detection – Find hats, scarves, belts, shoes, bags, gloves, etc.
- Fine-grained fashion details – Detect design features (pockets, buttons, zippers, collars, sleeves)
- Outfit composition analysis – Understand what items are being worn in an image
- Fashion inventory management – Automatically tag and categorize clothing inventory
- Style attribute extraction – Identify fashion design elements and characteristics
- Automated fashion annotation – Speed up product image labeling for catalog systems
Input/Output
- Input: Any RGB image of fashion items or clothing
- Content: Fashion items, clothing, accessories, outfits, fashion catalogs, product photos
- Output: List of detected fashion items and components
- Class label: What the fashion item is (e.g., "shirt", "pocket", "button", "collar")
- Bounding box: Location in image [x_min, y_min, x_max, y_max]
- Confidence score: How sure (0.0–1.0, where 1.0 = 100% sure)
- Fashion Categories Detected (46 total):
- Tops: shirt/blouse, t-shirt/sweatshirt, sweater, cardigan, jacket, vest, top
- Bottoms: pants, shorts, skirt
- Outerwear: coat, cape, jumpsuit, hood
- Accessories: glasses, hat, headband, tie, glove, watch, belt, leg warmer, tights, sock, shoe, bag, scarf, umbrella
- Details: collar, lapel, epaulette, sleeve, pocket, neckline, buckle, zipper, applique, bead, bow, flower, fringe, ribbon, rivet, ruffle, sequin, tassel
Performance
- Quality by Category:
- Large garments (shirts, pants, jackets): Excellent (90%+ confidence)
- Accessories (belts, scarves, bags): Very good (75-85% confidence)
- Fashion details (pockets, buttons, collars): Good (65-80% confidence)
- Small components (sequins, beads): Challenging (50-70% confidence)
- Multiple items: Good, handles full outfits well
- Advantages over General-Purpose Detectors:
- Recognizes 46 fashion-specific categories (general detectors only have 80 generic classes)
- Better at detecting small fashion details (buttons, zippers, pockets)
- Understands fashion context and components
- Domain-specific training improves accuracy on fashion items
Technical Details
Architecture | Vision Transformer (ViT-Small) Fine-Tuned for Fashion |
Key Innovation | Domain-specific fine-tuning enables detection of fashion-specific features (zippers, buttons, pockets, collars) that general detectors miss. |
Components | • Vision Transformer Encoder (ViT-Small): 12 layers, 12 attention heads, 384 hidden dimensions, 16×16 patch processing
• 100 Object Queries: Learnable embeddings that search for fashion items
• Fashion-Specific Prediction Heads: Classification (46 fashion classes) and bounding box regression (x, y, w, h) |
Training | • Base Model: YOLOS pre-trained on ImageNet-1k
• Fine-tuning: FashionPedia dataset (46.8k images, 46 fashion categories)
• Loss: Bipartite matching (Hungarian algorithm) + cross-entropy classification + L1 + Generalized IoU localization |
Requirements | • PyTorch (v1.7+)
• HuggingFace Transformers
• GPU recommended for inference |
Compliance & Provenance
Provider | Open-source |
Provider type | Specialized |
License | cc-by-4.0 |
EU AI Act risk class | Minimal Risk |
Art. 50 transparency | Not applicable |
Region availability | Available globally |
Training data summary | Pending — provider has not yet published per Art. 53(d) |
For more on how we classify models and mark outputs, see our AI Policy.