Install setup for the visualization

[JosephKKim]

Hi thanks for releasing interesting work,

I am following the instruction you've provided in README file, after following instructions from TEMOS, I found that path to blender in my linux system is /snap/blender/3135/2.93 When I set this path as a YOUR_BLENDER_PATH, this the first command doesn't work. There's blender file in /snap/blender/3135, so when I ran the command, it gives the blender version followed by OSError which says "GIT_REPOSITORY_PATH/motion-latent-diffusion/-m" could not be opened: No such file or directory" In this case, what should I do?

Thanks, Joseph.

[ChenFengYe]

Sorry for that wrong command, please check the following:

YOUR_BLENDER_PYTHON_PATH/python -m pip install -r prepare/requirements_render.txt

Please install these packages in prepare/requirements_render.txt for the specific python used in your blender.

Let me know if you still have more questions.

[JosephKKim]

Thanks for the quick reply!

Even after I typed the command you gave in the reply, it give the message which says

bash: /snap/blender/3135/2.93/python: Is a directory

Do you have any idea?

Thanks

[JosephKKim]

plus, can you please give me the numpy and pandas version of your environment?

[ChenFengYe]

Thanks for the quick reply!

Even after I typed the command you gave in the reply, it give the message which says

bash: /snap/blender/3135/2.93/python: Is a directory

Do you have any idea?

Thanks

You need to use the python executable file rather than a folder. It seems /snap/blender/3135/2.93/python is a folder. We will consider providing a docker image for running environment soon

[JosephKKim]

Thanks I think I found the executable file! I didn't know much about the blender.. BTW, I want to ask where can I find instructions for the datasets. Even running the demo file requires dataset but I cannot find instructions for the dataset in the README.

Thanks again for the fast reply!

[ChenFengYe]

You can visit https://github.com/EricGuo5513/HumanML3D for the dataset setup and then update HUMANML3D.ROOT and HUMANML3D.SPLIT_ROOT in https://github.com/ChenFengYe/motion-latent-diffusion/blob/main/configs/assets.yaml.

Actually, demo should not require the dataset for most tasks. We will try to update that part to remove this dependence and provide this dataset setup detail.

[JosephKKim]

Thanks for the reply! It would be really helpful if you can answer one more question. I've already posted issue on the HumanML3D, but didn't get a right answer, they wrote that we should process raw KIT dataset with the given ipynb file, but when I downloaded KIT dataset from the official website, it is npz file instead of npy file... I think that I misunderstood something... Can you tell me how to process the KIT file if you remember the process? Thanks

[ChenFengYe]

I do not suggest you download KIT from its official website, you can check the KIT provided by HumanML3D.

The below is from the readme of HumanML3D

For KIT-ML dataset, you could directly download [Here]. Due to the distribution policy of AMASS dataset, we are not allowed to distribute the data directly. We provide a series of script that could reproduce our HumanML3D dataset from AMASS dataset.

https://drive.google.com/drive/folders/1MnixfyGfujSP-4t8w_2QvjtTVpEKr97t

[JosephKKim]

Thanks for the so much help, After downloading and organizing HumanML3D dataset directory I see that demo is reading npy files from the dataset. But, I'm seeing the ValueError: not enough values to unpack (expected 2, got 0) .

While debugging I found that many motions goes into the bad_count and it skips the loop in dataset.py resulting empty new_name_list and length_list.

Empty new_name and lenght_list is true but, I think it's because of other reason. Is this a bug or am I doing something wrong again?

Thanks.

[ChenFengYe]

But, I'm seeing the ValueError: not enough values to unpack (expected 2, got 0

Could you provide the line and its file for this error? I think you might generate a wrong npy file.

[JosephKKim]

https://github.com/ChenFengYe/motion-latent-diffusion/blob/906b368b2c0864bdb51daebe3b42716c19b1734f/mld/data/humanml/data/dataset.py#L357

it is where the error occurs, and I am using HumanML3D dataset which is a bunch of npy files after unrar new_joint_vecs.rar file.

[ChenFengYe]

Before running the demo, could you please use the double-check process in humanml3d? https://github.com/EricGuo5513/HumanML3D/blob/main/cal_mean_variance.ipynb I guess your npy files might be broken.

You can also try to skip the dataset if you only want to run demo.

[JosephKKim]

Okay actually I am going to use this model for train/test. I ran cal_mean_variance.ipynb you referred to, encoutered an error processing the code below.

Am I using wrong npy file...? I thought I could skip this step after extracting the given rar file from the repository...

[ChenFengYe]

I believe your npy file is wrong. These npy files could be broken while your humanml generation. You could refer to https://github.com/EricGuo5513/HumanML3D/issues/3 for help.

[JosephKKim]

Hello, I fixed the npy file and then ran demo. 'python demo.py --cfg ./configs/config_mld_humanml3d.yaml --cfg_assets ./configs/assets.yaml --example ./demo/example.txt '

Seems like it created npy and txt file in samples_ directory,

After that I've followed all instructions from the visualization part, I want you to specify some details for me.

python -m fit --dir YOUR_NPY_FOLDER --save_folder TEMP_PLY_FOLDER --cuda

does YOUR_NPYFOLDER means the folder containing the npy files from the demo? In my case samples directory?

And TEMP_PLY_FOLDER is where I want to save ply files?

Added: What is the cuda option? Thanks

[ChenFengYe]

You are right. YOUR_NPY_FOLDER means the folder containing the npy files from the demo. TEMP_PLY_FOLDER means anywhere you want to save ply files. --cuda means using cuda to speed up this optimization.

[wangweikang1229]

我也出现了这样的问题 ValueError: not enough values to unpack( expected 2，got 0) 我重新提取了好几次humanml3d数据，但是问题没有解决。