Accuracy is very low for Hengshuang model

[alextarvo]

Hi,
I am trying to use your PCT code for my own research. After having issues training the PCT (i.e. the Hengshuang model), I ran the training example, as specified on your github page:

python train_partseg.py

The output:
iscander@muthr:~/pct/Point-Transformers$ python train_cls.py /home/iscander/pct/Point-Transformers/train_cls.py:44: UserWarning: The version_base parameter is not specified. Please specify a compatability version level, or None. Will assume defaults for version 1.1 @hydra.main(config_path='config', config_name='cls') /home/iscander/miniconda3/envs/cosarad/lib/python3.10/site-packages/hydra/_internal/defaults_list.py:251: UserWarning: In 'cls': Defaults list is missingself`. See https://hydra.cc/docs/1.2/upgrades/1.0_to_1.1/default_composition_order for more information
warnings.warn(msg, UserWarning)
/home/iscander/miniconda3/envs/cosarad/lib/python3.10/site-packages/hydra/_internal/hydra.py:119: UserWarning: Future Hydra versions will no longer change working directory at job runtime by default.
See https://hydra.cc/docs/1.2/upgrades/1.1_to_1.2/changes_to_job_working_dir/ for more information.
ret = run_job(
[2025-08-14 23:43:14,582][main][INFO] - Load dataset ...
The size of train data is 9843
The size of test data is 2468
[2025-08-14 23:43:14,862][main][INFO] - No existing model, starting training from scratch...
[2025-08-14 23:43:15,610][main][INFO] - Start training...
[2025-08-14 23:43:15,610][main][INFO] - Epoch 1 (1/200):
100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 616/616 [00:58<00:00, 10.53it/s]
[2025-08-14 23:44:14,150][main][INFO] - Train Instance Accuracy: 0.210024
100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 155/155 [00:12<00:00, 12.62it/s]
[2025-08-14 23:44:26,499][main][INFO] - Test Instance Accuracy: 0.115726, Class Accuracy: 0.076071
[2025-08-14 23:44:26,499][main][INFO] - Best Instance Accuracy: 0.115726, Class Accuracy: 0.076071
[2025-08-14 23:44:26,499][main][INFO] - Save model...
[2025-08-14 23:44:26,499][main][INFO] - Saving at best_model.pth
...
[2025-08-15 03:37:29,660][main][INFO] - Epoch 200 (200/200):
100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 616/616 [00:57<00:00, 10.66it/s]
[2025-08-15 03:38:27,535][main][INFO] - Train Instance Accuracy: 0.473688
100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 155/155 [00:12<00:00, 12.42it/s]
[2025-08-15 03:38:40,076][main][INFO] - Test Instance Accuracy: 0.606452, Class Accuracy: 0.498978
[2025-08-15 03:38:40,077][main][INFO] - Best Instance Accuracy: 0.626210, Class Accuracy: 0.524115
[2025-08-15 03:38:40,077][main][INFO] - End of training...

`

As you see, unfortunately, the accuracy is nowhere close to the advertised 90+%