ldz666666 / riddle Goto Github PK

Hi, I want to know if you use both PTI Tuning and Stitching Tuning to de-identify videos.

Thank you for your excellent work. I noticed that you actually use the w-code generated images to do the comparison of the recovered effect during the training and testing, why don't you use the recovered image to do the comparison with the original real image?

Hello author, how is the background maintenance in the paper achieved?It is not achievable during inference.

I know that MtCNN return 5 key points, but do you don't specify, if for Dlib results in the Table3, uses all 68 points. In #8 or document is no clear how many facial landmarks from Dlib are used.

Traceback (most recent call last):
File "coach_test.py", line 198, in
Coach=CoachTest(opts)
File "coach_test.py", line 41, in init
self.e4e=self.load_e4e()
File "coach_test.py", line 93, in load_e4e
ckpt = torch.load(self.opts.e4e_model_weights, map_location='cpu', _use_new_zipfile_serialization=False)
File "/project/liutaorong/anaconda3/envs/riddle/lib/python3.7/site-packages/torch/serialization.py", line 577, in load
with _open_zipfile_reader(opened_file) as opened_zipfile:
File "/project/liutaorong/anaconda3/envs/riddle/lib/python3.7/site-packages/torch/serialization.py", line 241, in init
super(_open_zipfile_reader, self).init(torch._C.PyTorchFileReader(name_or_buffer))
RuntimeError: [enforce fail at inline_container.cc:144] . PytorchStreamReader failed reading zip archive: not a ZIP archive

Hi,

For the facial landmarks and Bounding box distances reported in Table 3 of the main paper, the Euclidean norm is used as the distance metric, or another distance. I am trying to replicate these results, but the results obtained are not similar to those reported in the paper. Additionally, for Dlib landmarks uses all 68 points?

Hello! This is really interesting work! I am currently trying to run it, but encountered an issue during inference. It shows that the pre-trained segmentation model "./pretrained_models/79999_iter.pth" is missing or cannot be found ([Errno 2] No such file or directory: './pretrained_models/79999_iter.pth'). Upon checking the provided data and pre-trained model link, I did not find this specific pre-trained model mentioned.

This is great work; however, I noticed that currently the inference script only works with the FFHQ dataset. I am interested in using it to encrypt and decrypt custom images.

Could you please guide me on how to extend the functionality of the script to support custom images? I would appreciate any code snippets or documentation.

Hi,

Thank you for your amazing work. I would like to run de-identification on the CelebA-HQ dataset.
How can I do it? In particular, where can I find the inversion encoder that maps the original image to the latent space?

Thanks a lot.
Cheers,
Luigi

Hi everyone

anyone here who has run the full work?

Can I recover it by entering an anonymous picture instead of entering a w latent code?

Hello, I have some questions about Table 3 in the 4.5 Face Utility section of the paper.

The table compares the Bounding box distance and Landmark distance between the encrypted image and the original image. The distance used here is L1 distance or L2 distance, Landmark selects 5 or 68 points, and the distance in the table is the sum of distances or the average result. I did not see any relevant explanation in this paper.

I really hope to receive your answer.

Hi,I want to know how this embedding(invert_w_256.pt) was obtaine.