Benchmarks & Datasets
Collection
Evaluation benchmarks and training datasets for garment classification • 2 items • Updated
A balanced 10,000-sample training dataset for garment classification VLMs (v3).
JSONL with fields:
image: path to garment imagesource: annotation sourceresponse: ground-truth JSON with 9 fields (type, color, pattern, neckline, sleeve_length, closure, brand, size, defect_type)Used for supervised fine-tuning (SFT) and reinforcement learning (GRPO/GTPO) of vision-language models on structured garment attribute extraction.