not an issue - question on project_images

[johndpope]

Hi Justin,

maybe you can help me - my goal is to take a picture of myself and toonify myself.

I've compressed it into 3 steps

1. align the imges
2. project images
3. blend networks

this line - step 2.

python project_images.py --num-steps 500 aligned generated
which is going to go through the aligned photos / after being processed from raw has some training component. I'm not clear on why or what this is doing only that the generated folder is showing a bunch of blurred images.

WHY do we need to
Adding the images to tfrecords ...????
I want to take any image and find the associated npy.... do they all need to be added to tfrecords? and what does this actually do.

step 3 - just takes an array of npy file (the found vector of phot from latent space) and does the final step to make toon image. latent_dir.glob("*.npy")

https://gist.github.com/johndpope/55d4b8c33c3cac2c794373c4af7202ad

is there a faster / better way to just get the npy from an existing image?

[johndpope]

it seems like the logic for step2 could be collapsed around spitting out the npz/npy file
https://github.com/johndpope/stylegan2-ada/tree/digressions

python projector.py --outdir=out --target=targetimg.png \
    --network=https://nvlabs-fi-cdn.nvidia.com/stylegan2-ada/pretrained/ffhq.pkl

this creates 3 files in the out directory. targetimg.png / found latent image and the associated latent vector npy file.

[johndpope]

this projector code is old one from 2019
https://github.com/justinpinkney/stylegan2/blob/master/projector.py

you should abandon stylegan2 and use the stylegan2-ada repo.
I got it working with tensorflow2
https://github.com/johndpope/stylegan2-ada/blob/digressions/projector.py

UPDATE -
I tested the updated code / it also does 1000 steps / ticks to find the correspnding npy file.
When the toonify openapi do their processing - presumably they're not doing as many samples?