Rope3d Data generation problem

[Rickyyy-zh]

Hello:

when follow your prepare dataset md, I find that in scripts/gen_info_rope3d.py the train_infos is saved twice, both as training and validation label. Running training or evaluation code under this script might cause testing/validating on train data which leads to wrong result.
Another question is why re-distribute val and train data here? Using defalut training and val set will be helpful to re produce final result.

def main():
    rope3d_root = "data/rope3d"
    train_infos = generate_info_rope3d(rope3d_root, split='train')
    val_infos = generate_info_rope3d(rope3d_root, split='val')
    total_infos = train_infos + val_infos
    random.shuffle(total_infos)
    train_infos = total_infos[:int(0.7 * len(total_infos))]
    val_infos = total_infos[int(0.7 * len(total_infos)):]
    mmcv.dump(train_infos, './data/rope3d/rope3d_12hz_infos_train.pkl')
    # mmcv.dump(train_infos, './data/rope3d/rope3d_12hz_infos_val.pkl')    #<<=================line 223
    # should be:
    mmcv.dump(val_infos , './data/rope3d/rope3d_12hz_infos_val.pkl')

[M0NSTERCOCO]

Hello, the official download link for the Rope3D dataset is no longer working. Do you have a backup that you could provide? Thank you!