cy0307/b-hashgrid-instngp

Multiresolution hash grid (Instant-NGP)

A multiresolution hash-grid feature encoding + tiny MLP that fits a real photograph to high PSNR — the trick that made NeRF ~1000× faster.

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	differentiable image fitting
Training objective	Photometric L2 image fitting through a multiresolution hash-grid encoding + a tiny MLP.
Track	B · 3D & rendering
Notebook

Dataset

• Name: Real photograph (astronaut)
• Type: real — public-domain image
• Size / stats: 1 RGB photo resized to 96×96; 8-level hash grid (2^14 entries/level)
• Split: single image (overfit)
• Source: scikit-image data.astronaut() (NASA, public domain)

Training config

Adam (lr 1e-2), 1500 steps; 8-level hash grid (2¹⁴ entries × 2 feats/level) + 2-layer MLP; 96×96.

Evaluation results

metric	value	meaning
psnr (final)	64.25

Inference example

import torch
state = torch.load("hashgrid.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.

Overfits a single image; hash collisions limit the highest-frequency detail; PSNR is generous on smooth images.

Failure cases

Hash collisions cause speckle at the finest levels; fitting succeeds on one image but says nothing about others.

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.

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/B_hashgrid_instngp.ipynb --output run.ipynb
# optional: override training length, e.g.  STEPS=2000  (or EPISODES=600)  before running

Files

• figure.png
• hashgrid.pt
• metrics.json

License

Code & weights: MIT (this repository) — educational use encouraged.
Image: astronaut test image (NASA) — public domain, shipped with scikit-image.

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: Müller et al., Instant Neural Graphics Primitives (Instant-NGP), SIGGRAPH 2022.

Related assets

• 🚀 Live demos: https://huggingface.co/spaces/cy0307/ropedia-demos
• 🤗 All trained models + collection: https://huggingface.co/cy0307
• 📚 Course & all labs: https://chaoyue0307.github.io/ropedia-academy/ · Labs tab
• 💻 Source / notebooks: github.com/ChaoYue0307/ropedia-academy

---

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