Hi,

Command: python scripts/crop_align_vggface2_FFHQalign.py --input_dir /in --output_dir_ffhqalign /out --mode ffhq --crop_size 256

Gives error: Could not load library cudnn_ops_infer64_8.dll. Error code 126 Please make sure cudnn_ops_infer64_8.dll is in your library path!

cudnn_ops_infer64_8.dll is definitely there. I cloned my SimSwap environment, then "pip install"ed the extra requirements.

Any idea why this does not work?

Hi, I cannot unzip the zip files in Ubuntu 20.04, the file size seems OK, but got a lot of errors (only one image file is correctly extracted):

unzip VGGface2_HQ.zip
file #117124:  bad zipfile offset (local header sig):  9635728212

file VGGface2_HQ.zip
VGGface2_HQ.zip: DOS executable (COM, 0x8C-variant)

$ md5sum VGG*
b0b85056138ec65b545e8d43c17a1a4f  VGGface2_HQ.z01
9d0f8f07f5cda2ac7c32248ba54d220a  VGGface2_HQ.z02
f98596d310007c87fe45d2b1371aae24  VGGface2_HQ.z03
f025c8d036c027eb4ee5af173965d85f  VGGface2_HQ.z04
e5693cb420a648eb4630b4740f427c42  VGGface2_HQ.zip

This looks like an interesting dateset, but I am not sure how to open these files given their non-standard naming convention. I'm also not sure why one of them is a zip, while the others are not.

On Linux, could you go through the commands needed to open the files and organize them into the same file structure as the original VGGFace2?

/VGGFace2-HQ/VGGface2_HQ.zip
/VGGFace2-HQ/VGGface2_HQ.z01
/VGGFace2-HQ/VGGface2_HQ.z02
/VGGFace2-HQ/VGGface2_HQ.z03
/VGGFace2-HQ/VGGface2_HQ.z04

Thank you for your help.

Hi,

Thanks for great implementation. However, I am unable to extract these data. They are in different format. Could you show how to extract them to see image files?

Thanks

Sorry for the trivial question and ignorance, but it is often written where to copy the files in the simswap folder, in this case I don't find any information, okay in the main directory? thanks

the link of dataset VGGFace2-HQ not works i've loaded from https://academictorrents.com/details/535113b8395832f09121bc53ac85d7bc8ef6fa5b

The problem is that i don't know what's the right structure to decompress I have to create a VGGFace2 folder in the root folder? and data, meta?

Could you explain the right structure?

Thanks

After following the instructions to setup the environment, ./insightface_func/models, and ./experiments/pretrained_models...

Running the following command:

python scripts/crop_align_vggface2_FFHQalign.py --input_dir $DATAPATH$/VGGface2/train --output_dir_ffhqalign $ALIGN_OUTDIR$ --mode ffhq --crop_size 256

Results in the following output

input mean and std: 127.5 127.5
find model: ./insightface_func/models/antelope/glintr100.onnx recognition
find model: ./insightface_func/models/antelope/scrfd_10g_bnkps.onnx detection
set det-size: (320, 320)

  0%|                                                   | 0/500 [00:00<?, ?it/s]2021-12-20 16:21:17.029441668 [W:onnxruntime:, execution_frame.cc:811 VerifyOutputSizes] Expected shape from model of {800,10} does not match actual shape of {200,10} for output 500
2021-12-20 16:21:17.029507982 [W:onnxruntime:, execution_frame.cc:811 VerifyOutputSizes] Expected shape from model of {800,4} does not match actual shape of {200,4} for output 497
2021-12-20 16:21:17.029540840 [W:onnxruntime:, execution_frame.cc:811 VerifyOutputSizes] Expected shape from model of {800,1} does not match actual shape of {200,1} for output 494
2021-12-20 16:21:17.029912492 [W:onnxruntime:, execution_frame.cc:811 VerifyOutputSizes] Expected shape from model of {3200,10} does not match actual shape of {800,10} for output 477
2021-12-20 16:21:17.029973895 [W:onnxruntime:, execution_frame.cc:811 VerifyOutputSizes] Expected shape from model of {3200,4} does not match actual shape of {800,4} for output 474
2021-12-20 16:21:17.030020397 [W:onnxruntime:, execution_frame.cc:811 VerifyOutputSizes] Expected shape from model of {3200,1} does not match actual shape of {800,1} for output 471
...

It continues to throw warnings for hundreds of other weights, which I have not included here.

Is this expected behavior?

I unzipped the files on Google Drive, but it only creates a directory with 8631 classes for the training set. Is it possible for you to add the remaining 500 classes for VGGFace2 test set?

Thank you for sharing this resource!