XingangPan
Using custom images, creating a custom PTI results in loading error
Hi There 👋
Thanks a lot for the code. I've created a PTI using both the pti code inside this repo and the PTI repo you linked. The resulting checkpoints is smaller and in .pt file. If I try to load it into the ui by changing .pt -> pkl I got this error
File "/home/zuppif/miniconda3/envs/stylegan3/lib/python3.9/site-packages/anyio/to_thread.py", line 33, in run_sync
return await get_asynclib().run_sync_in_worker_thread(
File "/home/zuppif/miniconda3/envs/stylegan3/lib/python3.9/site-packages/anyio/_backends/_asyncio.py", line 877, in run_sync_in_worker_thread
return await future
File "/home/zuppif/miniconda3/envs/stylegan3/lib/python3.9/site-packages/anyio/_backends/_asyncio.py", line 807, in run
result = context.run(func, *args)
File "/home/zuppif/Documents/DragGAN/visualizer_drag_gradio.py", line 353, in on_change_pretrained_dropdown
init_images(global_state)
File "/home/zuppif/Documents/DragGAN/visualizer_drag_gradio.py", line 70, in init_images
state['renderer'].init_network(
File "/home/zuppif/Documents/DragGAN/viz/renderer.py", line 221, in init_network
G = self.get_network(pkl, 'G_ema')
File "/home/zuppif/Documents/DragGAN/viz/renderer.py", line 146, in get_network
raise data
viz.renderer.CapturedException: Traceback (most recent call last):
File "/home/zuppif/Documents/DragGAN/viz/renderer.py", line 138, in get_network
data = legacy.load_network_pkl(f)
File "/home/zuppif/Documents/DragGAN/legacy.py", line 23, in load_network_pkl
data = _LegacyUnpickler(f).load()
_pickle.UnpicklingError: A load persistent id instruction was encountered,
but no persistent_load function was specified.
For refence, the checkpoint dim is 127M while all yours are 348. My config for PTI is
## Pretrained models paths
e4e = '../pit/e4e_ffhq_encode.pt'
stylegan2_ada_ffhq = '../pti/ffhq.pkl'
style_clip_pretrained_mappers = ''
ir_se50 = '../pti/model_ir_se50.pth'
dlib = '../pti/shape_predictor_68_face_landmarks.dat'
## Dirs for output files
checkpoints_dir = './checkpoints'
embedding_base_dir = './embeddings'
styleclip_output_dir = './StyleCLIP_results'
experiments_output_dir = './output'
## Input info
### Input dir, where the images reside
input_data_path = '/home/zuppif/Documents/DragGAN/aligned_image/'
### Inversion identifier, used to keeping track of the inversion results. Both the latent code and the generator
input_data_id = 'barcelona'
## Keywords
pti_results_keyword = 'PTI'
e4e_results_keyword = 'e4e'
sg2_results_keyword = 'SG2'
sg2_plus_results_keyword = 'SG2_plus'
multi_id_model_type = 'multi_id'
## Edit directions
interfacegan_age = 'editings/interfacegan_directions/age.pt'
interfacegan_smile = 'editings/interfacegan_directions/smile.pt'
interfacegan_rotation = 'editings/interfacegan_directions/rotation.pt'
ffhq_pca = 'editings/ganspace_pca/ffhq_pca.pt'
Basically, the default. May I ask you how you created the checkpoints?
Thanks
Fra
I got it working, got the custom model, in pkl file but it is not generating the custom image from which this model was generated instead it is generating random human faces. do you have any idea?
okay, DONE 👍
I have successfully created custom pickle model and loaded w_pivot as well. I will soon share full step by step process.
damn, it is working like a charm.
Edit: uploaded a related step by step solution- https://youtu.be/viWiOC1Mikw
the .pkl model to choose in GUI is 'stylegan2_ada_ffhq = '../pti/ffhq.pkl' rename the 'ffhq.pkl' to 'stylegan2_ffhq.pkl' using the PTI will got the feature file under 'PTI/embedding/ .pt', load this .pt file as latent in GUI but the picture showed in GUI is not the same as what I sent into PTI not sure it is correct or not
the .pkl model to choose in GUI is 'stylegan2_ada_ffhq = '../pti/ffhq.pkl' rename the 'ffhq.pkl' to 'stylegan2_ffhq.pkl' using the PTI will got the feature file under 'PTI/embedding/ .pt', load this .pt file as latent in GUI but the picture showed in GUI is not the same as what I sent into PTI not sure it is correct or not
replace the ffhq.pkl['G_ema'] with the PTI output .pt file, the picture in GUI seems better, but still not the same
@FrancescoSaverioZuppichini @Antelope111
Hi I have created video step by step along with colab notebook https://youtu.be/viWiOC1Mikw
I will add instructions over here as well after some time. Any question most welcome 🤗
@FrancescoSaverioZuppichini @Antelope111
Hi I have created video step by step along with colab notebook https://youtu.be/viWiOC1Mikw
I will add instructions over here as well after some time. Any question most welcome hugs
I love you!
@Pawandeep-prog
How do I do this on Windows?
I was already able to install PTI and generate a .pt model, but how do I make a .pkl?
@Sipovec Use this colab notebook
You will get a function export_updated_pickle at the end of the notebook.just use it to convert to pickle.
Actually, I tried to embed the PTI into the DragGAN code, so that it doesn't need to be done separately. It kind of works, if you are interested, maybe you can have a look at my code (https://github.com/tianhaoxie/DragGAN_PTI). Any suggestions will be appreciated!
Reference
I initially used the pti inside the repo but I think there is a missing step to somehow convert that weights to the ones drag gan accepts. Would be curious to know how you did it
Actually, I tried to embed the PTI into the DragGAN code, so that it doesn't need to be done separately. It kind of works, if you are interested or are experienced in Gradio, maybe you can have a look at my code (https://github.com/tianhaoxie/DragGAN_PTI). Any suggestions will be appreciated!
thanks a lot for the other repo, I'll test it out. Would it make sense to have the whole code here as a separate branch. I think this would make it easier for people to help out given the more visibility of this repo
Yeah, that makes sense but I'm not sure how to do that correctly. If u can help, it will be appreciated. :)
its giving me a 'File Not Found' message when I try to run the second cell in "Save Latent space and Model Weights". im really new to coding so its probably a minor mistake but I can't seem to figure out how to fix
the .pkl model to choose in GUI is 'stylegan2_ada_ffhq = '../pti/ffhq.pkl' rename the 'ffhq.pkl' to 'stylegan2_ffhq.pkl' using the PTI will got the feature file under 'PTI/embedding/ .pt', load this .pt file as latent in GUI but the picture showed in GUI is not the same as what I sent into PTI not sure it is correct or not
replace the ffhq.pkl['G_ema'] with the PTI output .pt file, the picture in GUI seems better, but still not the same
I have the same problem, did you solve it?
