Hugging Face's logo

cy0307
/
a-motion-diffusion

PyTorch
ropedia-academy
educational
embodied-ai
from-scratch
reproducible
diffusion
Model card Files
xet
 Community
a-motion-diffusion / README.md
cy0307's picture
cy0307
Upload from Ropedia Academy
dff9293 verified
preview code
|
Raw
History Blame Contribute Delete
4.67 kB
metadata 
license: mit
library_name: pytorch
tags:
  - ropedia-academy
  - educational
  - embodied-ai
  - from-scratch
  - reproducible
  - diffusion

Motion diffusion (DDPM)

A denoising-diffusion model that generates motion trajectories — the MDM recipe, trained from scratch, sampled from EMA weights.

Trained from scratch in Ropedia Academy — an interactive, bilingual course on embodied & spatial AI. Educational model: small and quick to train; the value is the method and a reproducible pipeline, not a leaderboard score. Try it live in the Ropedia demos Space.

At a glance

Base model Trained from scratch (random initialization) — no pretrained base model.
Task human-motion generation
Training objective DDPM denoising objective (predict the added noise) over motion trajectories; sample from an EMA of the weights.
Track A · Human modeling
Notebook Open In Colab

Dataset

  • Name: Synthetic motion trajectories
  • Type: synthetic — procedural
  • Size / stats: 4,096 looping 2D trajectories, 32 timesteps × 2 = 64-D each (varied radius/phase/noise)
  • Split: train only (generative model)
  • Source: procedural

Training config

DDPM; Adam (lr 2e-4), 4000 steps, sequence length 32; EMA of weights for sampling. STEPS env-overridable.

Evaluation results

metric value meaning
loss (final) 0.159

figure

Inference example 

import torch
state = torch.load("denoiser.pt", map_location="cpu")   # this repo's checkpoint
# Rebuild the exact module from the lab notebook (see "Reproduce"), then:
# model.load_state_dict(state); model.eval()

Limitations

Educational scale. Trained quickly on CPU on small or synthetic data, so absolute numbers are not competitive with production systems — the value is the method and a reproducible pipeline. No large-scale data, no hyperparameter sweep, and no multi-seed variance is reported. Not for production use.

Failure cases

Too few sampling steps or no EMA → jittery/averaged motions; mode collapse with too little data.

Reproduce / train your own

One click: open the notebook in Colab → Runtime → GPU → Run all, then run its Publish to the Hugging Face Hub cell.

Open In Colab

From a shell:

git clone https://github.com/ChaoYue0307/ropedia-academy.git && cd ropedia-academy
pip install torch numpy matplotlib scikit-learn scikit-image gymnasium
jupyter nbconvert --to notebook --execute notebooks/training/A_motion_diffusion.ipynb --output run.ipynb
# optional: override training length, e.g.  STEPS=2000  (or EPISODES=600)  before running

Files

  • denoiser.pt
  • figure.png
  • metrics.json

License

Code & weights: MIT (this repository) — educational use encouraged.
Data: generated procedurally in the notebook — no external dataset.

Citation

If you use this model or the course materials, please cite:

@misc{ropedia_academy,
  title  = {Ropedia Academy: an interactive course on embodied & spatial AI},
  author = {Ropedia Academy},
  year   = {2026},
  howpublished = {\url{https://chaoyue0307.github.io/ropedia-academy/}}
}

Method / original work: Tevet et al., Human Motion Diffusion Model (MDM), ICLR 2023 (arXiv:2209.14916); Ho et al., DDPM, NeurIPS 2020.

Related assets

Part of the Ropedia Academy trained-model collection. Contributions & issues welcome on GitHub.