The objective of this repo is to simply overfit an MLP such that it takes coordinates as input and outputs colors. The data to overfit is provided in dataset/data
. This data originally comes from a mesh with vertex colors.
Until now, I've been unsuccessful in achieving this objective.
Folder | Description |
---|---|
dataset/data |
Contains the mesh data the needs to be overfit |
dataset/mesh_data.py |
Contains dataset class for training |
model/net_texture.py |
MLP with positional encoding |
trainer/train_overfit.py |
Training module |
runs | Experiment checkpoints and outputs |
Requirements are provided in requirements.txt
file. To install them, run
pip install -r requirements.txt
To train the MLP simply run
python trainer/train_overfit.py