A smart clothing recommender for my fashion-incompetent self built with FashionCLIP.
pip install -r requirements.txt
huggingface_hub==0.27.0
pip install 'rembg[cpu]'for using CPU- Downgrade to
transformations==4.42.4(see issue)
- FashionCLIP has a limit of 77 tokens, so keep the prompt to less than that
- Details from FashionCLIP finetuning on Farfetch dataset (unreleased)
- Dataset has mostly long captions --> longer prompts may be more effective
- Biased towards standard product images, i.e. centered on white background
- Generative try-on (VTON) feature --> CatVTON (github)
- Cache previously generated images to reduce inference frequency
- Intuitive UI for entering prompt + visualizing outfits
- Track clothing article usage frequency
- Add laundry (i.e. remove worn clothes from recommendation)
- Use weather report and emotion prediction for informing outfit (GradedRec for reference on traversing FashionCLIP latent space)
- FashionCLIP (github)
- Might replace with Marqo-FashionLCIP
- Outfit Transformer (github)
- VTON Research List 1 and VTON Research List 2