Awesome

U-RED: Unsupervised 3D Shape Retrieval and Deformation for Partial Point Clouds

Yan Di*, Chenyangguang Zhang*, Ruida Zhang*, Fabian Manhardt, Yongzhi Su, Jason Rambach, Didier Stricker, Xiangyang Ji, Federico Tombari

ICCV 2023

Data download and preprocessing details

Please follow joint_learning_retrieval_deformation to prepare the dataset and reorder the directory like:

your_base_dir/
    data_aabb_all_models/
    data_aabb_constraints_keypoint/
    dis_mat/
    generated_datasplits/
    partnet_rgb_masks_chair/
    partnet_rgb_masks_storagefurniture/
    partnet_rgb_masks_table/

Setup

pip install -r requirements.txt

Configs

Modify the config files in folder 'config'. Complete the value occupied by 'xxx' including "base_dir" and "log_path" in training config and "dm_model_path", "re_model_path", "base_dir" and "log_path" in testing config. During testing phase, keep "dm_model_path" and "re_model_path" exactly the same as the trained model. If you want to change the category, just fix "category" into "storagefurniture" or "table".

When running engine/vis.py script, it needs to first set up https://github.com/mhsung/libigl-renderer, and then modify Line 160 of dataset/dataset_utils.py with the corresponding path.

Train

CUDA_VISIBLE_DEVICES=0 python engine/train.py config/config_train_chair.json

Test

CUDA_VISIBLE_DEVICES=0 python engine/test.py config/config_test_chair.json

Visualization

CUDA_VISIBLE_DEVICES=0 python engine/vis.py config/config_vis_chair.json