- HumanML3D includes 14,616 motion samples across diverse actions (walking, dancing, sports) and 44,970 text annotations.
- Data includes skeletal joint positions, rotations, and fine-grained textual descriptions.

<img src="https://external-content.duckduckgo.com/iu/?u=https%3A%2F%2Fproduction-media.paperswithcode.com%2Fdatasets%2F446194c5-ce59-43eb-b4cb-570a7a4d0cd9.png&f=1&nofb=1&ipt=b2edbe3251cab88e26a7f9d4e765c811b2cc890dc2ace7f7456baeca076b115b&ipo=images" alt="description" style="width:800px; height:600px;" />

The provided dataset contains the following components:

- 1. `motions` Folder: Contains `.npy` files, each representing a sequence of body poses. Each file has a shape of `(T, N, d)`, where:
  - `T`: Number of frames in the sequence (varies across sequences).
  - `N`: Number of joints in the body (22 in this case).
  - `d`: Dimension of each joint (3D coordinates: `x`, `y`, `z`).

- 2. `texts` Folder: Contains `.npy` files, each providing **3 textual descriptions** of the corresponding motion sequence. Each description is accompanied by part-of-speech (POS) tags for every word in the description. Example: "a person jump hop to the right#a/DET person/NOUN jump/NOUN hop/NOUN to/ADP the/DET right/NOUN#"

- 3. File Lists
    - **`all.txt`**: List of all motion files in the dataset.
    - **`train.txt`**: List of motion files for training.
    - **`val.txt`**: List of motion files for validation.
    - **`test.txt`**: List of motion files for testing.
   - **`Challenge_Human_Motion_Generation_Notebook.ipynb`**: Notebook for exploring and visualization of sequences.
    - **`sampleSubmission.csv`**: a sample submission file in the correct format.