Sygil-Dev/muse-maskgit-pytorch issues and pull requests

#83 - ValueError: the image you passed in is not of the correct dimensions

Issue - State: open - Opened by Aabie 10 months ago

#82 - Small QOL when using the validation_folder_at_end_of_epoch argument, we now append the epoch number to the filename so we can identify at what epoch the image was saved.

Pull Request - State: closed - Opened by ZeroCool940711 about 1 year ago

#81 - Flax support?

Issue - State: open - Opened by nRuaif over 1 year ago - 3 comments

#80 - Any results available?

Issue - State: open - Opened by AhmedGamal411 over 1 year ago - 2 comments

#79 - idk if this works, but here's a small clip implementation

Pull Request - State: open - Opened by korakoe over 1 year ago - 3 comments

#78 - Add support for loading multiple datasets

Pull Request - State: closed - Opened by korakoe over 1 year ago - 1 comment

#77 - Fix minor issue

Pull Request - State: closed - Opened by korakoe over 1 year ago - 3 comments

#76 - cleanup + embed saving + loading

Pull Request - State: closed - Opened by korakoe over 1 year ago - 2 comments

#75 - Early precomputation implementation

Pull Request - State: closed - Opened by korakoe over 1 year ago - 1 comment

#74 - Small QOL when using the validation_folder_at_end_of_epoch argument. We now append the epoch number to the filename so we can identify at what epoch the image was saved.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#73 - Added some utility code to reconstruct a whole folder with the VAE, since this code is used on multiple places now I decided to move it to the utils.py script to reuse it.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#72 - Added even more support for the EMA model.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#71 - Fixed the VAE validation images being sent too many times to wandb during validation causing network issues because of it hitting the api rate limit.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#70 - Partially restored the ability to use the EMA model for training on the VAE trainer.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#69 - The VAE validation images are now saved as individual image pairs rather than a single grid of images.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#68 - Account for potentially corrupt images in dataset, and disable strict loading

Pull Request - State: closed - Opened by korakoe over 1 year ago - 1 comment

#67 - Changes all the print statements to use `accelerator.print` for compatibility with latest accelerate library and colab.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#66 - Fixed missing wandb dependency.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#65 - Added some missing args to the Arguments class.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#64 - Added several QOL improvements to the infer_vae.py script.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#63 - Merge pull request #62 from Sygil-Dev/dev

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#62 - Merge dev to main.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#61 - Fixed the dataset not converting properly the images to RGBA when using 4 channels for training on the maskgit.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#60 - Fixed the infer_vae.py script so it would retain the alpha channel whens saving the grid if we are using `--channels 4`, previosly it was generating properly the reconstructed image with alpha channel but the grid was converting the image to RGB instead of RGBA which removed the alpha channel, now it properly handles both RGB and RGBA images.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#59 - Fixed the alpha channel not working properly when initializing a new VAE with 4 channels.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#58 - [pre-commit.ci] pre-commit autoupdate

Pull Request - State: open - Opened by pre-commit-ci[bot] over 1 year ago

#57 - Added support for both 3 and 4 channels images.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#56 - [pre-commit.ci] pre-commit autoupdate

Pull Request - State: closed - Opened by pre-commit-ci[bot] over 1 year ago

#55 - Added arguments to customize the number of channels, layers and discr_layers for the vae.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#54 - Added a couple of lines to print the total number of parameters the model has after it has been loaded for the vae and maskgit training scripts as well as the infer_vae.py

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#53 - The logging_dir cli arg will now use "args.results_dir/logs" if you leave it blank, it should reduce the need to manually add it every time.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#52 - Create sem-version-release.yml

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#51 - Fixed error when using a validation_image_scale value higher than 1.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#50 - [pre-commit.ci] pre-commit autoupdate

Pull Request - State: closed - Opened by pre-commit-ci[bot] over 1 year ago

#49 - Added option to customize the timesteps value for the maskgit training.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#48 - Improved the epoch counter to also take into account the number of epochs trained before based on the current number of steps when resuming from a checkpoint.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#47 - Improved some prints on the console for the maskgit training script.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#46 - Disabled most warnings on the t5 script to keep the console clean and nice.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#45 - Update README.md

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#44 - Merge pull request #43 from Sygil-Dev/dev

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#43 - Merge Dev to Main

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#42 - Update README.md

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#41 - Removed duplicated code for the maskgit training script.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#40 - Merge pull request #39 from Sygil-Dev/dev

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#39 - Merge dev to main

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#38 - Fixed funding.yml location and added more of my accounts that can be used for donations.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#37 - Merge pull request #36 from Sygil-Dev/dev

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#36 - Merge dev to main.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#35 - Create FUNDING.xml and several fixes.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#34 - Changed the pretrained "pretrained=True" to be "weights=torchvision.models.VGG16_Weights.DEFAULT" so it is compatible with latest version of torchvision and also so we always get the latest version of the weights.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#33 - Fixed issue where the training would stop early than intended.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#32 - VQVAE loading fix

Pull Request - State: closed - Opened by neggles over 1 year ago

#31 - Update .pre-commit-config.yaml

Pull Request - State: closed - Opened by neggles over 1 year ago

#30 - add VQVAE from PaintMind

Pull Request - State: closed - Opened by neggles over 1 year ago - 2 comments

#29 - Small fix to the step counter on the vae trainer when saving the model to disk, the print was showing the counter as having 1 step less than it actually had.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#28 - Fixed missing arguments on the train_muse_vae.py script.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#27 - Merge main to dev.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#26 - Merge dev to main.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#25 - actions: fix pre-commit again

Pull Request - State: closed - Opened by neggles over 1 year ago

#24 - actions: fix pre-commit

Pull Request - State: closed - Opened by neggles over 1 year ago

#23 - Small fixes made by the pre-commit hook.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#22 - Add pre-commit config and enforce code formatting + sanity checks

Pull Request - State: closed - Opened by neggles over 1 year ago

#21 - Added option to specify the project name, run name and user/organization to use for trackers such as wandb.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#20 - Implemented the attention types

Pull Request - State: closed - Opened by korakoe over 1 year ago

#19 - Several more fixes and improvements.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#18 - Reverted changes to the progress bar description which made it so it was making a new line on each steps which was not intended to happen on the progress bar.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#17 - Several fixes, improvements and QOL.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#16 - Moved message for the validation images so its shown before the progress bar for readability.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#15 - Freeze the VAE before passing to the transformer

Pull Request - State: closed - Opened by korakoe over 1 year ago

#14 - Added several fixes and QOL options.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#13 - fix dataset loading twice

Pull Request - State: closed - Opened by korakoe over 1 year ago

#12 - Idk what happened but conflicts are resolved for me now

Pull Request - State: closed - Opened by korakoe over 1 year ago

#11 - Added random_crop, hf_split_name and renamed skip_arrow to no_cache for clarity as well as other fixes.

Pull Request - State: closed - Opened by ZeroCool940711 over 1 year ago

#10 - Prevent possible re-instantiation

Pull Request - State: closed - Opened by korakoe over 1 year ago

#9 - vqgan_vae_taming: fix broken einops map

Pull Request - State: closed - Opened by neggles over 1 year ago

#8 - update tpu-vm.env

Pull Request - State: closed - Opened by korakoe over 1 year ago - 1 comment

#7 - update tpu-vm.env

Pull Request - State: closed - Opened by korakoe over 1 year ago

#6 - implement epoch system for muse scheduler

Pull Request - State: closed - Opened by korakoe over 1 year ago

#5 - Birch-san attn implementations

Pull Request - State: closed - Opened by korakoe over 1 year ago - 1 comment

#4 - update tpu-vm.env

Pull Request - State: closed - Opened by neggles over 1 year ago

#3 - Implement proper scheduling with epoch system

Pull Request - State: closed - Opened by korakoe over 1 year ago

#2 - implement Random cropping

Pull Request - State: closed - Opened by korakoe over 1 year ago - 1 comment

#1 - Resolve Validation dataloader StopIteration

Pull Request - State: closed - Opened by korakoe over 1 year ago

GitHub / Sygil-Dev/muse-maskgit-pytorch issues and pull requests