StreetTryOn, the new in-the-wild Virtual Try-On dataset, consists of 12,364 and 2,089 street person images for training and validation, respectively. It is derived from the large fashion retrieval dataset DeepFashion2, from which we filter out over 90% of DeepFashion2 images that are infeasible for try-on tasks (e.g., non-frontal view, large occlusion, dark environment, etc.). Combining with the garment and person images in VITON-HD, we obtain a comprehensive suite of in-domain and cross-domain try-on tasks that have garment and person inputs from various sources, including Shop2Model, Model2Model, Shop2Street, and Street2Street.

• 12,364 training images and 2,089 test images
• Tests included:
  • Shop2Model
  • Model2Model
  • Shop2Street
  • Model2Street
  • Street2Street