Hi! Your work is so meaningful! Nowadays I'm trying to caculate the FID score using your released model trained with SHHQ-1.0 (~40K, SHHQ-1.0_sg2_512.pk), then I use it to generate 50K images and caculate FID with SHHQ-1.0, but my FID score (7.12) is much higher than that you reported (3.68). Could you share more details about your reported FID? What images you use to caculate FID? Is there a test set using for FID? By the way, could you release your training settings of 40K model and 230K model, it is not clear that if they are trained with similar iters or epochs.

run:
!python alignment.py --image-folder /content/input --output-folder /content/output. ,

but I got

Traceback (most recent call last):
  File "alignment.py", line 221, in <module>
    run(cmd_args)
  File "alignment.py", line 144, in run
    comb, segmentation, bg, ori_img = human_seg.run(image,None)  #mybg) 
  File "/content/StyleGAN-Human/PP_HumanSeg/deploy/infer.py", line 100, in run
    processed_imgs, ori_shapes = self.preprocess(img)
  File "/content/StyleGAN-Human/PP_HumanSeg/deploy/infer.py", line 92, in preprocess
    processed_img = self.compose(img)[0]
  File "/usr/local/lib/python3.7/dist-packages/paddleseg/transforms/transforms.py", line 56, in __call__
    if 'img' not in data.keys():
AttributeError: 'numpy.ndarray' object has no attribute 'keys'

My environment

• colab pro.
• input folder: /content/input; only one image, of which size is (512, 1024), PIL image.
• To solve module error, !pip install paddleseg paddlepaddle

and I believe, the problem is here, self.compose = T.Compose(self.cfg.transforms).
self.cfg.transforms wants to get dictionary having a key, 'img'.
but I cant get though it.

Plz help.

I tried two methods in order to find the latent code with your pre-trained model, stylegan_human_v2_1024.pkl and e4e_w+.pt

1 - I started with some seeds (e.g. 0, 4000, 4050, 4020, 5500, 5000, 10) and generated some images. Then I treated those images as real images and tried to find their latent representations using ("python run_pti.py"), I used the default config in "pti_configs" folder. In some cases, the face, and in others, the pattern of the cloth is distorted (please see the following pictures)

2 - I downloaded the sample images from your GitHub and did the same procedure. The aforementioned problems are more noticeable. I got distorted faces and patterns.

I was wondering if you have any advice for these cases.

Thank you for releasing the codes for PTI encoding of whole body images. I will like to know if there's a way to speed up the process of PTI encoding.

Thanks for releasing Inversion script. Projection is working fine but the result is poor. We can try hyper parameter tuning. But you have given another option e4e. ( w+) . It is failing "key error"

File "/StyleGAN_Human/pti/training/coaches/base_coach.py", line 136, in initilize_e4e opts = ckpt['opts'] KeyError: 'opts'

I tried printing the key values in ckpt ( loading the e4ew+.pt) . There is no such key "opts". Could you please tell us what can be done to get e4e based inversion ? I expect it will give better output than projection.

Hi, thanks for your great job and the released codes.

When I tried to edit the images I found that there are some seeds should be fed into the model, e.g. [60948,60965,61174,61210,61511,61598,61610] #bottom -> long. What are the meanings of these seeds and how to obtain the corresponding seeds if I would like to perform other kinds of editing?

Thank you for your help.

Dear StyleGAN-Human team,

Thank you for sharing this great work, I really like it.

Could you upload the FID of each StyleGAN model? Have you try to use the projected gan or vision-aided-gan to improve the generation quality?

Thank you for your help.

Best Wishes,

Zongze

Hello, I am trying to run the code of insetgan on Google Colab and I've run in this issue.

Previously, the code was running without errors.

Has anyone run in the same issue? How could this problem be solved?

This is the output of insetgan.py:

Setting up [LPIPS] perceptual loss: trunk [alex], v[0.1], spatial [off]
/usr/local/lib/python3.7/dist-packages/torchvision/models/_utils.py:209: UserWarning: The parameter 'pretrained' is deprecated since 0.13 and will be removed in 0.15, please use 'weights' instead.
  f"The parameter '{pretrained_param}' is deprecated since 0.13 and will be removed in 0.15, "
/usr/local/lib/python3.7/dist-packages/torchvision/models/_utils.py:223: UserWarning: Arguments other than a weight enum or `None` for 'weights' are deprecated since 0.13 and will be removed in 0.15. The current behavior is equivalent to passing `weights=AlexNet_Weights.IMAGENET1K_V1`. You can also use `weights=AlexNet_Weights.DEFAULT` to get the most up-to-date weights.
  warnings.warn(msg)
Loading model from: /usr/local/lib/python3.7/dist-packages/lpips/weights/v0.1/alex.pth
Traceback (most recent call last):
  File "insetgan.py", line 398, in <module>
    main()
  File "/usr/local/lib/python3.7/dist-packages/click/core.py", line 829, in __call__
    return self.main(*args, **kwargs)
  File "/usr/local/lib/python3.7/dist-packages/click/core.py", line 782, in main
    rv = self.invoke(ctx)
  File "/usr/local/lib/python3.7/dist-packages/click/core.py", line 1066, in invoke
    return ctx.invoke(self.callback, **ctx.params)
  File "/usr/local/lib/python3.7/dist-packages/click/core.py", line 610, in invoke
    return callback(*args, **kwargs)
  File "/usr/local/lib/python3.7/dist-packages/click/decorators.py", line 21, in new_func
    return f(get_current_context(), *args, **kwargs)
  File "insetgan.py", line 372, in main
    _, body_crop, _ = insgan.detect_face_dlib(body_img)
  File "insetgan.py", line 110, in detect_face_dlib
    output_size=256)
  File "/content/StyleGAN-Human/utils/face_alignment.py", line 40, in align_face_for_insetgan
    lm, face_rect = get_landmark(img, detector, predictor)
  File "/content/StyleGAN-Human/utils/face_alignment.py", line 18, in get_landmark
    dets = detector(img, 1)
RuntimeError: Error while calling cudnnConvolutionBiasActivationForward( context(), &alpha1, descriptor(data), data.device(), (const cudnnFilterDescriptor_t)filter_handle, filters.device(), (const cudnnConvolutionDescriptor_t)conv_handle, (cudnnConvolutionFwdAlgo_t)forward_algo, forward_workspace, forward_workspace_size_in_bytes, &alpha2, out_desc, out, descriptor(biases), biases.device(), identity_activation_descriptor(), out_desc, out) in file /tmp/pip-install-7f5va9na/dlib_dd01df0de7e94b8893437a0dcaab1525/dlib/cuda/cudnn_dlibapi.cpp:1237. code: 9, reason: CUDNN_STATUS_NOT_SUPPORTED

I'm trying to additionally train a model for StyleGAN3 using my own images (512x512 resolution).

python train.py --outdir=training_results/sg3/ --cfg=stylegan3-r --gpus=8 --batch=32 --gamma=12.4 \
    --mirror=1 --aug=noaug --data=data/OrgData/ --square=True --snap=250

When I run the prepared script, I get the following error:

Traceback (most recent call last):
  File "train_stylegan_human.py", line 293, in <module>
    main() # pylint: disable=no-value-for-parameter
  File "/opt/conda/lib/python3.8/site-packages/click/core.py", line 1128, in __call__
    return self.main(*args, **kwargs)
  File "/opt/conda/lib/python3.8/site-packages/click/core.py", line 1053, in main
    rv = self.invoke(ctx)
  File "/opt/conda/lib/python3.8/site-packages/click/core.py", line 1395, in invoke
    return ctx.invoke(self.callback, **ctx.params)
  File "/opt/conda/lib/python3.8/site-packages/click/core.py", line 754, in invoke
    return __callback(*args, **kwargs)
  File "train_stylegan_human.py", line 288, in main
    launch_training(c=c, desc=desc, outdir=opts.outdir, dry_run=opts.dry_run)
  File "train_stylegan_human.py", line 98, in launch_training
    subprocess_fn(rank=0, c=c, temp_dir=temp_dir)
  File "train_stylegan_human.py", line 49, in subprocess_fn
    training_loop.training_loop(rank=rank, **c)
  File "/home/rd/img_studio/source/submodules/StyleGAN3/training/training_loop.py", line 163, in training_loop
    misc.copy_params_and_buffers(resume_data[name], module, require_all=False)
  File "/home/rd/img_studio/source/submodules/StyleGAN3/torch_utils/misc.py", line 164, in copy_params_and_buffers
    tensor.copy_(src_tensors[name].detach()).requires_grad_(tensor.requires_grad)
RuntimeError: The size of tensor a (512) must match the size of tensor b (1024) at non-singleton dimension 1

I set --cbase=16384 referring to the StyleGAN3 issue, but the result did not change.

Can someone please tell me how to deal with this?

Thanks for your great work. I got access to SHHQ dataset recently. I have a few doubts.

1. Alignment of images : I could see dataset images with many different backgrounds. I just want to know which is better? Batch processing of alignment of images and create a new dataset. Use this new dataset for training

or including alignment / other preprocessing as a part of training. Just want to know the best practice? which one you followed ?

2. In the paper, You have mentioned Guassian blur and changing to uniform background. May i assume your alignment script will do both?

3. How to use the segmention mask you have provided with the dataset ( for alignment ) in this training?

I find there are two functions in your code G.mapping and G.synthesis. But if I want to 18 different codes in w+ space without using the affine transformation in G.synthesis. How to achieve that? Thanks a lot!!

@stylegan-human, the FID I am getting for stylegan_human_v2_1024.pkl seems higher than what is reported in the paper.

I did the following two methods for FID computation:

1. Generated 10 K images and calculated FID using pytorch-fid to get FID of ~59
2. Use the calc_metrics.py from the stylegan2-ada-pytorch repo and calculate the fid50k_full to be 50.0

Both of these calculations are done with respect to the SHHQ-1.0 40K images.

Just curious what dataset was used as reference to compute the FID on your generated images! Is it possible to share the FID computation code for reproducibility? Thanks for your consideration.

How I can Prepare my Dataset with Labels to Train this Model? I lable my data set as provided here in the PyTorch Docs

Here a Example frome my CSV: "Image_Name""test.png", "Male":True, "Hair_Black":False, "Hair_Brown":True

The original StyleGAN Lib use only Integer Lables. How I can get the benefits in Learning from my Dataset? Thanks

Thank you for the amazing job you are doing! You have achieved amazing results and the generated images look fantastic!

Please release the training code so we can train the model further (and on top of ) on our own time for personal needs. I see a lot of people asking for it - please make it available for us . Thank you !

Hi, Thanks for your amazing job,

From the style mixing results, the middle mixing will influence the cloth and the id appearance at the same time, I am curious about whether you try to point out the mixing way only changing the style of cloth when keeping the identity.