Skip to content

The multi-modal universe of fast-fashion: the Visuelle 2.0 benchmark

License

Notifications You must be signed in to change notification settings

HumaticsLAB/visuelle2.0-code

Repository files navigation

The multi-modal universe of fast-fashion: the Visuelle 2.0 benchmark

The official pytorch implementation of models discussed in The multi-modal universe of fast-fashion: the Visuelle 2.0 benchmark paper.

Accepted as oral at the 5th Workshop on Computer Vision for Fashion, Art, and Design @ CVPR22

Installation

We suggest the use of VirtualEnv.

git clone https://github.com/HumaticsLAB/visuelle2.0-code.git
cd visuelle2.0-code
virtualenv mmrnn_venv # If you don't have virtualenv, you can install it by using "pip install virtualenv"
source mmrnn_venv/bin/activate # or mmrnn_venv\Scripts\activate.bat # If you're running on Windows

pip install torch==1.8.2 torchvision==0.9.2 torchaudio==0.8.2 --extra-index-url https://download.pytorch.org/whl/lts/1.8/cu111
pip install pytorch-lightning==1.6.5
pip install pandas numpy opencv-python Pillow scikit-image scikit-learn scipy tqdm fairseq wandb statsmodels

# To deactivate the virtual env simply execute "deactivate"

Dataset

VISUELLE2 dataset is publicly available here. Please download and extract it inside the root folder. A more detailed description of the dataset can be found in the project webpage.

Run Naive and Simple Exponential Smoothing baselines

# Naive Method
python forecast_stat.py --method naive --use_teacher_forcing 1  #SO-fore2−1
python forecast_stat.py --method naive --use_teacher_forcing 0  #SO-fore2−10

# Simple Exponential Smoothing Method
python forecast_stat.py --method ses --use_teacher_forcing 1  #SO-fore2−1
python forecast_stat.py --method ses --use_teacher_forcing 0  #SO-fore2−10

Run SO-fore2−1

python train_dl.py --task_mode 0
python forecast_dl.py --task_mode 0 --ckpt_path <ckpt_path>

Run SO-fore2−10

python train_dl.py --task_mode 1
python forecast_dl.py --task_mode 1 --ckpt_path <ckpt_path>

Run Demand SO-fore

python train_dl.py --new_product 1
python forecast_dl.py --new_product 1 --ckpt_path <ckpt_path>

Model zoo

You can find the pretrained models here.

Citation

If you use the VISUELLE2.0 dataset or this particular implementation, please cite the following paper:

@inproceedings{skenderi2022multi,
  title={The multi-modal universe of fast-fashion: the Visuelle 2.0 benchmark},
  author={Skenderi, Geri and Joppi, Christian and Denitto, Matteo and Scarpa, Berniero and Cristani, Marco},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={2241--2246},
  year={2022}
}

About

The multi-modal universe of fast-fashion: the Visuelle 2.0 benchmark

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published