/siren-tf2

Unofficial SIRENs implementation using Tensorflow 2.

Primary LanguagePythonMIT LicenseMIT

Siren and FilmSiren TF 2 : Implicit Neural Representations with Periodic Activation Functions

The unofficial Tensorflow 2 implementation of the paper Implicit Neural Representations with Periodic Activation Functions. Please note that, this repo tested with image fitting experiments.

Paper | Official PyTorch Implementation

Get Started

To start working with this repository please install Python packages using:

conda env create --file setup/environment.yaml

Training

python main.py --train --input_image samples/leaves.jpg  --output_dir results/leaves --n_epochs 1000

Testing

python main.py --input_image samples/leaves.jpg  --output_dir results/leaves/

Results

3 layers, 256x256 image size training 1000 epoch mean square errors.

Image_file Baseline TF
istanbul_airport.jpg 0.0020 0.0056
face.jpg 0.0021 0.0061
stone_nsm.jpg 0.0011 0.0016
leaves.jpg 0.0014 0.0037

References