yzliu

> @yzliu567 Hi, you need to revise the `rand_light: [-0.9,0.9,-0.3,0.8,-0.1,0.7,-0.4]` parameter in the config file. > The corresponding light sampling code is at `gan2shape/model.py` Line 752-762. This is also described...

Hi. Sorry we don't have this model on hand. You can follow `encoder4editing/train_sh.txt` to train one from the given e4e pretrained model and our agesmileglasses GAN model. You may need...

Hi, the order should be age-glasses-smile. See [this](https://github.com/yzliu567/sc-gan/blob/main/dataset_ffhq.py#L53) for the attribute order. You can also uncomment code [here](https://github.com/yzliu567/sc-gan/blob/main/train_mmd.py#L244) to visualize the image and corresponding mode vector to have a check.

Sorry for the confusion, you are right.

Yes. You just need all the attributes' values, the uniform sampling result and changing the mode number in the architecture. But too many attributes may not work well since the...

Hello. You need to first invert the original dataset into latent vectors (70000 x 18 x 512) using official e4e model. Then compute the distances to the boundaries you pick...

We used 1 single Titan Xp to finetune the model for 30k to 50k iterations, which takes about 12 hrs. Please refer to our paper for more details.

It is likely caused by too many attributes. As I mentioned above, when the attributes are mutually dependent and the number of attributes is large, the uniform sampling method may...