normally noise:

uniform noise:

lower traingular kernel:

random noise:

visualizing nca's in python/jupyter notebook:

• 256, 256, 3 dim normally distributed tensors with value in range [0, 1] are used as the starting board / playground
• kernel/convolutional filter of size 3, 3, 3 with values in range [0, 1]
• activation function is defaulted to sigmoid but can be changed to anything i.e np.tanh ...

random noise example:

git clone "https://github.com/hashirkz/neural_cellular_automata" <directory to save to e.x /mnt/c/users/<urname>/documents>
code <directory u saved to>
pip3 install -r 'requirements.txt'

# 1. edit the kernel to your preference keeping its dimensions as 3, 3, 3
# 2. edit the starting playground or leave as random noise
# 3. run the notebook and view the gifs in ./gifs

1. currently not sure why the imgs gradually lose color as they evolve/simulate could be activation function / hyperparameters
2. imageio.v3 is not working correctly not my fault the gifs produced flicker the first frame of the gif occasionally