problem about training the track transformer

[KaiTang98]

Hello, thank you for presenting this interesting work. I am attempting to reproduce the training based on the Libero dataset.

However, I've noticed that when training the track transformer, the train_dataset_list and val1_dataset_list obtained through:

train_dataset_list = glob(os.path.join(root_dir, f"{suite_name}//train/")) val1_dataset_list = glob(os.path.join(root_dir, f"{suite_name}//val/"))

are empty. I'm wondering if the data preprocessing stage requires additional processing of the data obtained from preprocess_libero.py? Many thanks!

[KaiTang98]

I checked the dataset inside linero_goal as an example. The dataset structure is shown as follows. -- libero_goal ------ open_the_middle_drawer.... ----------images ----------videos ----------demo_0.hdf5 ----------demo_1.hdf5 ----------....hdf5 ----------env_meta.json

[KaiTang98]

I realized that I just need to create these two folders by myself... Now I can train the model...😂😂

[caichuang0415]

I realized that I just need to create these two folders by myself... Now I can train the model...😂😂

Thank you very much I have just come with this problem as well 😂😂

[yolo01826]

@KaiTang98 训练的结果怎么样，能够完成任务吗

[ahadjawaid]

I found the script you are suppose to run here is the command:

python3 scripts/split_libero_dataset.py --folder data/atm_libero --train_ratio 0.9

[AlvinWen428]

Thank you very much for pointing this out. The "train" and "val" folders are generated by scripts/split_libero_dataset.py. Sorry for missing this important step in README. I have updated it.