All notable changes to this project will be documented in this file.
The format is based on Keep a Changelog,
and this project adheres to Semantic Versioning.
1.2.0 - 10/02/2022
- New custom saving time; default is 25 minutes - DeepFake ENG ITA
- Added random blur to shadow augmentation
- Splittable random shadows
- Added a handful of checks to indicate that the dataset is in zip or pak form when the functions used require a bulk dataset
- true face power only avaible when gan power > 0
- wrong def. cpu cap
- Other random fixes
1.1.0 - 29/12/2021
- 'random_hsv_power' from Official fork - seranus
- New 'force_full_preview' to force to do not separate dst, src and pred views in different frames - randomfaker
- Refactored two pass splitting it into 3 mode: None, face, face + mask - randomfaker
- Updated shadow augmentation splitting it in: None, src, dst, all - DeepFake ENG ITA
- Update requirements-colab.txt
1.0.0 - 10/12/2021
We created this fork from several other forks of DeepFaceLab. Many features of this fork comes mainly from JH's fork.
- Web UI for training preview
- Random color training option
- Background Power training option
- MS-SSIM loss training option
- GAN label smoothing and label noise options
- MS-SSIM+L1 loss function, based on "Loss Functions for Image Restoration with Neural Networks"
- Autobackup options:
- Session name
- ISO Timestamps (instead of numbered)
- Max number of backups to keep (use "0" for unlimited)
- New sample degradation options (only affects input, similar to random warp):
- Random noise (gaussian/laplace/poisson)
- Random blur (gaussian/motion)
- Random jpeg compression
- Random downsampling
- New "warped" preview(s): Shows the input samples with any/all distortions.
- FaceSwap-Aug in the color transfer modes
- Custom face types
- External configuration files by Cioscos aka DeepFake ENG ITA
- use --auto_gen_config CLI param to auto generate config. file or resume its configuration
- use --config_training_file CLI param external configuration file override
- Tensorboard support by JanFschr aka randomfaker
- AMP training updates - DeepFake ENG ITA & randomfaker
- shadow augmentation (needs testing to see if it can generalise well) - randomfaker
- filename labels by Ognjen aka JesterX aka seranus
- zip faceset support - randomfaker
- exposed new configuration parameters (cpu, lr, preview samples)
- Added pre-sharpen into the merger. It helps the model to fit better to the target face. Idea taken from DeepFaceLive
- Added two pass option into the merger. It processes the generated face twice. Idea taken from DeepFaceLive