The zip file ./shared_data/fine-grained_texts.zip (link) stores FineHumanML3D textual descriptions generated from HumanML3D with our prompt on GPT-3.5-turbo-0301. Due to copyright issues, please follow the instructions of HumanML3D to prepare motion data by yourself.
The data entries look like below. Each .txt file corresponds to one HumanML3D motion file with the same name. There are usually multiple descriptions in one .txt file, and each one of them describes the same motion from different aspects.
./texts_all_code_v2_marked_texts_tagged
├── 000000.txt
├── 000001.txt
├── 000002.txt
├── 000003.txt
├── ...
├── M000000.txt
├── M000001.txt
├── M000002.txt
├── M000003.txt
├── ...
When you use these fine-grained descriptions, please bear in mind that our fine-grained texts do not correspond to all the coarse-grained descriptions in HumanML3D, for we deleted some texts due to problematic responses from GPT-3.5-turbo-0301. In such situations, we simply leave the corresponding lines in the .txt files blank.
For example, while the original 010146.txt in HumanML3D has two coarse-grained descriptions in two lines, the 010146.txt in FineHumanML3D also has two lines, but the first line is left blank because GPT-3.5-turbo-0301 did not return valid responses, resulting in only one fine-grained description for this motion.
If you are using the FineHumanML3D dataset, please consider citing both our paper and the HumanML3D paper:
@inproceedings{li-feng-2024-motion-generation,
title = "Motion Generation from Fine-grained Textual Descriptions",
author = "Li, Kunhang and Feng, Yansong",
booktitle = "Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024)",
month = may,
year = "2024",
address = "Torino, Italy",
publisher = "ELRA and ICCL",
url = "https://aclanthology.org/2024.lrec-main.1016",
pages = "11625--11641"
}
@InProceedings{Guo_2022_CVPR,
author = {Guo, Chuan and Zou, Shihao and Zuo, Xinxin and Wang, Sen and Ji, Wei and Li, Xingyu and Cheng, Li},
title = {Generating Diverse and Natural 3D Human Motions From Text},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2022},
pages = {5152-5161}
}