Why use --base interfacegan

[dinusha94]

In the readme why it is used as --base interfacegan, except --base ours? to reproduce the results

[yxuhan]

Hi, --base ours means the semantic direction obtained by averaging the instance-specific semantic direction of lots of random images. Recall our main contribution is to introduce instance-aware latent space search into editing in the latent space, not a way to compute the base semantic direction, so we just use the base direction from interfacegan as the default option.

[dinusha94]

Hi, Thanks for the answer, Can you tell me that the following procedure is correct, I need to change the hair color and/or style of a face image

Steps I followed so far

1. I trained a binary classifier model with hair color images, say black and non-black images from celebahq dataset (lets say classifier as hair)
2. use python train_attr_level_direction.py --n_images 500 --attr hair to create direction file hair.npy file
3. use python edit_single_attr.py --seed 0 --step 0.5 --n_steps 4 --dataset celebhq --base interfacegan --attr hair --save_path test_env.jpg

Thanks in advance