search

IIT-PAVIS / DiffAssemble

Official repository for "DiffAssemble: A Unified Graph-Diffusion Model for 2D and 3D Reassembly" accepted at CVPR2024
48 stars 4 forks source link

One question about Breaking bad #4

Open pluto2D opened 2 weeks ago

pluto2D commented 2 weeks ago

Dear authors,

Do you plan to release the train_beakingbad.sh and it's ckpt?

Stefa1994 commented 1 week ago

Hi and thanks for your interest in our work! To train the Breaking Bad dataset, you can use the following command in the command line:

python puzzle_diff/train_3d.py --inference_ratio 10 --batch_size 8 --steps 300 --num_workers 10 --noise_weight 0 --predict_xstart True  --max_epochs 1000 --backbone vn_dgcnn --max_num_part 20

We plan to release the checkpoint, so stay tuned!