Loaded an empty dataset

[Leocoding0971]

Hi, thank you for the great work on this project!

I'm trying to run the following command:
python -m sample.generate --model_path ./save/humanml_trans_dec_512_bert/model000200000.pt --text_prompt "the person walked forward and is picking up his toolbox."

However, I get this error:
AssertionError: You loaded an empty dataset, it is probably because your data dir has only texts and no motions.
To train and evaluate MDM you should get the FULL data as described in the README file.

I've downloaded the dataset as mentioned in the README, but I'm not sure what exactly is missing. Could you please clarify what files are required in the data directory for sampling?

Thanks in advance for your help!

[noah6868]

@Leocoding0971 hello i met the same question, have you solved it?

[Leocoding0971]

Sorry guy, i haven't solved it yet. If you can run the program successfully, please let me know. Thank you!

…

---Original--- From: ***@***.***> Date: Wed, Jun 18, 2025 23:17 PM To: ***@***.***>; Cc: ***@***.******@***.***>; Subject: Re: [GuyTevet/motion-diffusion-model] Loaded an empty dataset (Issue#249) noah6868 left a comment (GuyTevet/motion-diffusion-model#249) @Leocoding0971 hello i met the same question, have you solved it? — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: ***@***.***>

[noah6868]

@Leocoding0971 hello i finally solve this problem, i think you download the dataset through the author's link( first methodDownload HumanML3D , but the author's dataset not include the [texts] folder, i share this folder in this link, just download it and put it in /MDM/dataset/HumanML3D/ , and you will solve this problem : )

[Leocoding0971]

@Leocoding0971 hello i finally solve this problem, i think you download the dataset through the author's link( first methodDownload HumanML3D , but the author's dataset not include the [texts] folder, i share this folder in this link, just download it and put it in /MDM/dataset/HumanML3D/ , and you will solve this problem : )

Thanks for the solution! It did work for me.

[ntg7creation]

im having the same problem

@Leocoding0971 hello i finally solve this problem, i think you download the dataset through the author's link( first methodDownload HumanML3D , but the author's dataset not include the [texts] folder, i share this folder in this link, just download it and put it in /MDM/dataset/HumanML3D/ , and you will solve this problem : )

And the link you give still requires a Request from the owner so I will wait and see if it works

[Leocoding0971]

im having the same problem

@Leocoding0971 hello i finally solve this problem, i think you download the dataset through the author's link( first methodDownload HumanML3D , but the author's dataset not include the [texts] folder, i share this folder in this link, just download it and put it in /MDM/dataset/HumanML3D/ , and you will solve this problem : )

And the link you give still requires a Request from the owner so I will wait and see if it works

Hi, you can download the [texts] file from "kaggle.com", it provides HumanML3D dataset which including a text file.

[noah6868]

im having the same problem

@Leocoding0971 hello i finally solve this problem, i think you download the dataset through the author's link( first methodDownload HumanML3D , but the author's dataset not include the [texts] folder, i share this folder in this link, just download it and put it in /MDM/dataset/HumanML3D/ , and you will solve this problem : )

And the link you give still requires a Request from the owner so I will wait and see if it works

sorry, this is the time i use google drive, you can also download from other site. And i have agree the request and i have changed the file permission to everyone.
不好意思，第一次使用谷歌云盘，你也可以从其他位置下载这个 [text.zip]，我已经通过了你们的请求，也把文件权限修改成所有人了

[Trinitas]

I got the same problem a while ago. I think the problem was that the "texts" was in a zip file

[OrangeJuFan]

Hi, thank you for the great work on this project!

I'm trying to run the following command: python -m sample.generate --model_path ./save/humanml_trans_dec_512_bert/model000200000.pt --text_prompt "the person walked forward and is picking up his toolbox."

However, I get this error: AssertionError: You loaded an empty dataset, it is probably because your data dir has only texts and no motions. To train and evaluate MDM you should get the FULL data as described in the README file.

I've downloaded the dataset as mentioned in the README, but I'm not sure what exactly is missing. Could you please clarify what files are required in the data directory for sampling?

Thanks in advance for your help!

I encountered the same issue as well. The bug isn't caused by a missing 'texts' folder as suggested in the comments section; rather, it is triggered by this specific line of code.

When the use_cache value is set to True, the code attempts to read data from a cache file. When set to False, it reads data directly from our server's disk. Since this is the first time running the training code, the cache file is currently empty. Therefore, for the initial run, you must change use_cache to False to avoid this error. Once the first run is complete, you can switch it back to True.