4.1. TSegNet 학습

1. Centroid Prediction Module 학습

   • train_configs/tsegnet.py 파일에서 run_tooth_segmentation_module = False로 설정.

   start_train.py \\
    --model_name "tsegnet" \\
    --config_path "train_configs/tsegnet.py" \\
    --experiment_name "tsegnet_centroid" \\
    --input_data_dir_path "path/for/preprocessed_data" \\
    --train_data_split_txt_path "base_name_train_fold.txt" \\
    --val_data_split_txt_path "base_name_val_fold.txt"
   
   

2. Tooth Segmentation Module 학습

   • train_configs/tsegnet.py 파일에서 run_tooth_segmentation_module = True로 설정.
   • pretrained_centroid_model_path를 이전 단계에서 저장된 체크포인트 경로로 설정.

   start_train.py \\
    --model_name "tsegnet" \\
    --config_path "train_configs/tsegnet.py" \\
    --experiment_name "tsegnet_segmentation" \\
    --input_data_dir_path "path/for/preprocessed_data" \\
    --train_data_split_txt_path "base_name_train_fold.txt" \\
    --val_data_split_txt_path "base_name_val_fold.txt"
   
   

4.2. PointNet 또는 PointNet++ 학습

start_train.py \\
 --model_name "pointnet" \\ # 또는 pointnetpp
 --config_path "train_configs/pointnet.py" \\ # pointnetpp.py로 변경 가능
 --experiment_name "pointnet_experiment" \\
 --input_data_dir_path "path/for/preprocessed_data" \\
 --train_data_split_txt_path "base_name_train_fold.txt" \\
 --val_data_split_txt_path "base_name_val_fold.txt"

4.3. TGNet 학습

• Farthest Point Sampling (FPS):

  start_train.py \\
   --model_name "tgnet_fps" \\
   --config_path "train_configs/tgnet_fps.py" \\
   --experiment_name "tgnet_fps" \\
   --input_data_dir_path "path/for/preprocessed_data" \\
   --train_data_split_txt_path "base_name_train_fold.txt" \\
   --val_data_split_txt_path "base_name_val_fold.txt"
  
  

• Boundary Aware Point Sampling (BDL):

  • train_configs/tgnet_bdl.py 파일에서 다음 4개 경로 수정:
    • original_data_obj_path
    • original_data_json_path
    • bdl_cache_path
    • load_ckpt_path

  start_train.py \\
   --model_name "tgnet_bdl" \\
   --config_path "train_configs/tgnet_bdl.py" \\
   --experiment_name "tgnet_bdl" \\
   --input_data_dir_path "path/for/preprocessed_data" \\
   --train_data_split_txt_path "base_name_train_fold.txt" \\
   --val_data_split_txt_path "base_name_val_fold.txt"
  
  

5. 모델 추론

5.1. 일반 모델 추론 (PointNet, TSegNet 등)

python start_inference.py \\
 --input_dir_path obj/file/parent/path \\
 --split_txt_path base_name_test_fold.txt \\
 --save_path path/to/save/results \\
 --model_name pointnet \\ # 또는 pointnetpp, tsegnet 등
 --checkpoint_path your/model/checkpoint/path

5.2. TGNet 추론

python start_inference.py \\
 --input_dir_path obj/file/parent/path \\
 --split_txt_path base_name_test_fold.txt \\
 --save_path path/to/save/results \\
 --model_name tgnet_fps \\
 --checkpoint_path your/tgnet_fps/checkpoint/path \\
 --checkpoint_path_bdl your/tgnet_bdl/checkpoint/path

6. 평가 및 시각화

• 단일 예제 평가 및 시각화:

python eval_visualize_results.py \\
 --mesh_path path/to/obj_file \\
 --gt_json_path path/to/gt_json_file \\
 --pred_json_path path/to/predicted_json_file

7. 결과 분석