Thanks for your great work and inference code. do you plan to opensource your training code and training data, if yes, can you provide a approximate time. Thanks in advance
We plan to make the training code available on acceptance. For the training data, please check this repo and use the available reconstruction method on the MEAD dataset.
Thanks for your great work and inference code. do you plan to opensource your training code and training data, if yes, can you provide a approximate time. Thanks in advance