FAME-ViL: Multi-Tasking Vision-Language Model for Heterogeneous Fashion Tasks

Updates

• 😍 (21/03/2023) Our FAME-ViL is selected as a highlight paper at CVPR 2023! (Top 2.5% of 9155 submissions)
• 😊 (12/03/2023) Code released!

Our trained model is available at Google Drive.

Please refer to FashionViL repo for the dataset preparation.

Test on FashionIQ

python mmf_cli/run.py \
config=projects/fashionclip/configs/mtl_wa.yaml \
model=fashionclip \
datasets=fashioniq \
checkpoint.resume_file=save/backup_ckpts/fashionclip_512.pth \
run_type=test \
model_config.fashionclip.adapter_config.bottleneck=512