- Allow selecting folder as well as multiple images
- Lower default threshold and update helptext
- bugfix: detector error when using all aligned faces
- Standardize output folder name
- Remove old face filter
- plugins.extract.pipeline: Expose plugins directly
- Change `is_aligned` from plugin level to ExtractMedia level
- Allow extract pipeline to take faceswap aligned images
- Add ability for recognition plugins to accept aligned faces as input
- Add face filter to recognition plugin
- Move extractor pipeline IO ops to own class
- alignments.py
- Add typed dicts for imported alignments
- Explicitly check for presence of thumb value in alignments dict
- linting
- detected_face.py
- Typing
- Linting
- Legacy support for pre-aligned face
- Update dependencies to new property names
* Minimum requirements to tf2.3.
- Handle upgrades for Windows users with tf2.2 installed by Pip
- Handle windows upgrade from pip tf2.2
- Explicitly install Cuda for Conda installs
* Update tensorflow errors api reference
* Suppress AutoGraph warning messages
* Update GUI Stats to work with tf2.3
* Fix live graph for tf2.3
* DSSIMObjective - autoGraph bugfix
* Update Travis test
* Extract
- Implement aligner re-feeding
- Add extract type to pipeline.ExtractMedia
- Add pose annotation to debug
* Convert
- implement centering
- remove usage of feed and reference face properties
- Remove distributed option from convert
- Force update of alignments file on legacy receive
* Train
- Resize preview image to model output size
- Force legacy centering if centering does not exist in model's state file
- Enable training on legacy face sets
* Alignments Tool
- Update draw to include head/pose
- Remove DFL drop + linting
- Remove remove-frames job
- remove align-eyes option
- Update legacy masks to new extract type
- Exit if attempting to merge version 1.0 alignments files with version 2.0 alignments files
- Re-generate thumbnails on legacy upgrade
* Mask Tool
- Update for new extract + bugfix full frame
* Manual Tool
- Update to new extraction method
- Disable legacy alignments,
- extract box bugfix
- extract faces - size to 512 and center on head
* Preview Tool
- Display based on model centering
* Sort Tool
- Use alignments for sort by face
* lib.aligner
- Add Pose Class
- Add AlignedFace Class
- center _MEAN_FACE on x
- Add meta information with versioning to alignments file
- lib.aligner.get_align_matrix to use landmarks not face
- Refactor aligned faces in lib.faces_detect
* lib.logger
- larger file log padding
* lib.config
- Fix global changeable_items
* lib.face_filter
- Use new extracted face images
* lib.image
- bump thumbnail default size to 96px
* Core Updates
- Remove lib.utils.keras_backend_quiet and replace with get_backend() where relevant
- Document lib.gpu_stats and lib.sys_info
- Remove call to GPUStats.is_plaidml from convert and replace with get_backend()
- lib.gui.menu - typofix
* Update Dependencies
Bump Tensorflow Version Check
* Port extraction to tf2
* Add custom import finder for loading Keras or tf.keras depending on backend
* Add `tensorflow` to KerasFinder search path
* Basic TF2 training running
* model.initializers - docstring fix
* Fix and pass tests for tf2
* Replace Keras backend tests with faceswap backend tests
* Initial optimizers update
* Monkey patch tf.keras optimizer
* Remove custom Adam Optimizers and Memory Saving Gradients
* Remove multi-gpu option. Add Distribution to cli
* plugins.train.model._base: Add Mirror, Central and Default distribution strategies
* Update tensorboard kwargs for tf2
* Penalized Loss - Fix for TF2 and AMD
* Fix syntax for tf2.1
* requirements typo fix
* Explicit None for clipnorm if using a distribution strategy
* Fix penalized loss for distribution strategies
* Update Dlight
* typo fix
* Pin to TF2.2
* setup.py - Install tensorflow from pip if not available in Conda
* Add reduction options and set default for mirrored distribution strategy
* Explicitly use default strategy rather than nullcontext
* lib.model.backup_restore documentation
* Remove mirrored strategy reduction method and default based on OS
* Initial restructure - training
* Remove PingPong
Start model.base refactor
* Model saving and resuming enabled
* More tidying up of model.base
* Enable backup and snapshotting
* Re-enable state file
Remove loss names from state file
Fix print loss function
Set snapshot iterations correctly
* Revert original model to Keras Model structure rather than custom layer
Output full model and sub model summary
Change NNBlocks to callables rather than custom keras layers
* Apply custom Conv2D layer
* Finalize NNBlock restructure
Update Dfaker blocks
* Fix reloading model under a different distribution strategy
* Pass command line arguments through to trainer
* Remove training_opts from model and reference params directly
* Tidy up model __init__
* Re-enable tensorboard logging
Suppress "Model Not Compiled" warning
* Fix timelapse
* lib.model.nnblocks - Bugfix residual block
Port dfaker
bugfix original
* dfl-h128 ported
* DFL SAE ported
* IAE Ported
* dlight ported
* port lightweight
* realface ported
* unbalanced ported
* villain ported
* lib.cli.args - Update Batchsize + move allow_growth to config
* Remove output shape definition
Get image sizes per side rather than globally
* Strip mask input from encoder
* Fix learn mask and output learned mask to preview
* Trigger Allow Growth prior to setting strategy
* Fix GUI Graphing
* GUI - Display batchsize correctly + fix training graphs
* Fix penalized loss
* Enable mixed precision training
* Update analysis displayed batch to match input
* Penalized Loss - Multi-GPU Fix
* Fix all losses for TF2
* Fix Reflect Padding
* Allow different input size for each side of the model
* Fix conv-aware initialization on reload
* Switch allow_growth order
* Move mixed_precision to cli
* Remove distrubution strategies
* Compile penalized loss sub-function into LossContainer
* Bump default save interval to 250
Generate preview on first iteration but don't save
Fix iterations to start at 1 instead of 0
Remove training deprecation warnings
Bump some scripts.train loglevels
* Add ability to refresh preview on demand on pop-up window
* Enable refresh of training preview from GUI
* Fix Convert
Debug logging in Initializers
* Fix Preview Tool
* Update Legacy TF1 weights to TF2
Catch stats error on loading stats with missing logs
* lib.gui.popup_configure - Make more responsive + document
* Multiple Outputs supported in trainer
Original Model - Mask output bugfix
* Make universal inference model for convert
Remove scaling from penalized mask loss (now handled at input to y_true)
* Fix inference model to work properly with all models
* Fix multi-scale output for convert
* Fix clipnorm issue with distribution strategies
Edit error message on OOM
* Update plaidml losses
* Add missing file
* Disable gmsd loss for plaidnl
* PlaidML - Basic training working
* clipnorm rewriting for mixed-precision
* Inference model creation bugfixes
* Remove debug code
* Bugfix: Default clipnorm to 1.0
* Remove all mask inputs from training code
* Remove mask inputs from convert
* GUI - Analysis Tab - Docstrings
* Fix rate in totals row
* lib.gui - Only update display pages if they have focus
* Save the model on first iteration
* plaidml - Fix SSIM loss with penalized loss
* tools.alignments - Remove manual and fix jobs
* GUI - Remove case formatting on help text
* gui MultiSelect custom widget - Set default values on init
* vgg_face2 - Move to plugins.extract.recognition and use plugins._base base class
cli - Add global GPU Exclude Option
tools.sort - Use global GPU Exlude option for backend
lib.model.session - Exclude all GPUs when running in CPU mode
lib.cli.launcher - Set backend to CPU mode when all GPUs excluded
* Cascade excluded devices to GPU Stats
* Explicit GPU selection for Train and Convert
* Reduce Tensorflow Min GPU Multiprocessor Count to 4
* remove compat.v1 code from extract
* Force TF to skip mixed precision compatibility check if GPUs have been filtered
* Add notes to config for non-working AMD losses
* Rasie error if forcing extract to CPU mode
* Fix loading of legace dfl-sae weights + dfl-sae typo fix
* Remove unused requirements
Update sphinx requirements
Fix broken rst file locations
* docs: lib.gui.display
* clipnorm amd condition check
* documentation - gui.display_analysis
* Documentation - gui.popup_configure
* Documentation - lib.logger
* Documentation - lib.model.initializers
* Documentation - lib.model.layers
* Documentation - lib.model.losses
* Documentation - lib.model.nn_blocks
* Documetation - lib.model.normalization
* Documentation - lib.model.session
* Documentation - lib.plaidml_stats
* Documentation: lib.training_data
* Documentation: lib.utils
* Documentation: plugins.train.model._base
* GUI Stats: prevent stats from using GPU
* Documentation - Original Model
* Documentation: plugins.model.trainer._base
* linting
* unit tests: initializers + losses
* unit tests: nn_blocks
* bugfix - Exclude gpu devices in train, not include
* Enable Exclude-Gpus in Extract
* Enable exclude gpus in tools
* Disallow multiple plugin types in a single model folder
* Automatically add exclude_gpus argument in for cpu backends
* Cpu backend fixes
* Relax optimizer test threshold
* Default Train settings - Set mask to Extended
* Update Extractor cli help text
Update to Python 3.8
* Fix FAN to run on CPU
* lib.plaidml_tools - typofix
* Linux installer - check for curl
* linux installer - typo fix
- Lint simple_tests.py
- Only reformat alignments file if it exists otherwise change filename
- Update legacy alignments to new format at all stages
- faces_detect.Mask.from_dict - logging format fix
- convert.py fix otf for new pipeline
- cli.py - Add note that masks not used. Revert convert masks
- faces_detect.py - Revert non-extract code
- Add .p and .pickle extensions for serializer
- plugins/extract revert some changes
- scripts/fsmedia - Revert code changes
- Pipeline - cleanup
- Consistant alpha channel stripping (fixes single-process)
- Store landmarks as numpy array
- Code attribution
- Normalize feed face and reference face to 0.0 - 1.0 in convert
- Lock in mask VRAM sized
- Add documentation to plugin_loader
- Update alignments tool to work with new format
- Remove legacy update hashes
- Remove legacy job from alignment-tools
- Remove legacy landmark rotation
- Add rotate face method to plugins/extract/detect
- Update travis test for new alignments extension
- Alignments format to .fsa
- Remove serializer option from alignments-tool
- Auto update legacy format alignment files to new format
* Move image utils to lib.image
* Add .pylintrc file
* Remove some cv2 pylint ignores
* TrainingData: Load images from disk in batches
* TrainingData: get_landmarks to batch
* TrainingData: transform and flip to batches
* TrainingData: Optimize color augmentation
* TrainingData: Optimize target and random_warp
* TrainingData - Convert _get_closest_match for batching
* TrainingData: Warp To Landmarks optimized
* Save models to threadpoolexecutor
* Move stack_images, Rename ImageManipulation. ImageAugmentation Docstrings
* Masks: Set dtype and threshold for lib.masks based on input face
* Docstrings and Documentation
* requirements.txt: - Pin opencv to 4.1.1 (fixes cv2-dnn error)
* lib.face_detect.DetectedFace: change LandmarksXY to landmarks_xy. Add left, right, top, bottom attributes
* lib.model.session: Session manager for loading models into different graphs (for Nvidia + CPU)
* plugins.extract._base: New parent class for all extract plugins
* plugins.extract.pipeline. Remove MultiProcessing. Dynamically limit batchsize for Nvidia cards. Remove loglevel input
* S3FD + FAN plugins. Standardise to Keras version for all backends
* Standardize all extract plugins to new threaded codebase
* Documentation. Start implementing Numpy style docstrings for Sphinx Documentation
* Remove s3fd_amd. Change convert OTF to expect DetectedFace object
* faces_detect - clean up and documentation
* Remove PoolProcess
* Migrate manual tool to new extract workflow
* Remove AMD specific extractor code from cli and plugins
* Sort tool to new extract workflow
* Remove multiprocessing from project
* Remove multiprocessing queues from QueueManager
* Remove multiprocessing support from logger
* Move face_filter to new extraction pipeline
* Alignments landmarksXY > landmarks_xy and legacy handling
* Intercept get_backend for sphinx doc build
# Add Sphinx documentation
Deprecation Warning: Rotation in Extract
Deprecation Warning: Multiple models within a single folder
Error Handling: Useful message for training size assertion error
Error Handling: Useful message for fewer images than batch size
Bugfix: BoundingBox object sometimes not available inside spawned process
Bugfix: Fully disable keypress monitor for GUI
Bugfix: Preview - Handle missing alignments file
Config changes:
- Separate plugin defaults into their own files
- Move mask_type to global training config
- Add ability to pass in custom config files
* Implement extraction pipeline
* Face filter to vgg_face. Resume partial model downloads
* On-the-fly conversion to extraction pipeline
* Move git model ids from get_model to model definition
* model_refactor (#571)
* original model to new structure
* IAE model to new structure
* OriginalHiRes to new structure
* Fix trainer for different resolutions
* Initial config implementation
* Configparse library added
* improved training data loader
* dfaker model working
* Add logging to training functions
* Non blocking input for cli training
* Add error handling to threads. Add non-mp queues to queue_handler
* Improved Model Building and NNMeta
* refactor lib/models
* training refactor. DFL H128 model Implementation
* Dfaker - use hashes
* Move timelapse. Remove perceptual loss arg
* Update INSTALL.md. Add logger formatting. Update Dfaker training
* DFL h128 partially ported
* Add mask to dfaker (#573)
* Remove old models. Add mask to dfaker
* dfl mask. Make masks selectable in config (#575)
* DFL H128 Mask. Mask type selectable in config.
* remove gan_v2_2
* Creating Input Size config for models
Creating Input Size config for models
Will be used downstream in converters.
Also name change of image_shape to input_shape to clarify ( for future models with potentially different output_shapes)
* Add mask loss options to config
* MTCNN options to config.ini. Remove GAN config. Update USAGE.md
* Add sliders for numerical values in GUI
* Add config plugins menu to gui. Validate config
* Only backup model if loss has dropped. Get training working again
* bugfixes
* Standardise loss printing
* GUI idle cpu fixes. Graph loss fix.
* mutli-gpu logging bugfix
* Merge branch 'staging' into train_refactor
* backup state file
* Crash protection: Only backup if both total losses have dropped
* Port OriginalHiRes_RC4 to train_refactor (OriginalHiRes)
* Load and save model structure with weights
* Slight code update
* Improve config loader. Add subpixel opt to all models. Config to state
* Show samples... wrong input
* Remove AE topology. Add input/output shapes to State
* Port original_villain (birb/VillainGuy) model to faceswap
* Add plugin info to GUI config pages
* Load input shape from state. IAE Config options.
* Fix transform_kwargs.
Coverage to ratio.
Bugfix mask detection
* Suppress keras userwarnings.
Automate zoom.
Coverage_ratio to model def.
* Consolidation of converters & refactor (#574)
* Consolidation of converters & refactor
Initial Upload of alpha
Items
- consolidate convert_mased & convert_adjust into one converter
-add average color adjust to convert_masked
-allow mask transition blur size to be a fixed integer of pixels and a fraction of the facial mask size
-allow erosion/dilation size to be a fixed integer of pixels and a fraction of the facial mask size
-eliminate redundant type conversions to avoid multiple round-off errors
-refactor loops for vectorization/speed
-reorganize for clarity & style changes
TODO
- bug/issues with warping the new face onto a transparent old image...use a cleanup mask for now
- issues with mask border giving black ring at zero erosion .. investigate
- remove GAN ??
- test enlargment factors of umeyama standard face .. match to coverage factor
- make enlargment factor a model parameter
- remove convert_adjusted and referencing code when finished
* Update Convert_Masked.py
default blur size of 2 to match original...
description of enlargement tests
breakout matrxi scaling into def
* Enlargment scale as a cli parameter
* Update cli.py
* dynamic interpolation algorithm
Compute x & y scale factors from the affine matrix on the fly by QR decomp.
Choose interpolation alogrithm for the affine warp based on an upsample or downsample for each image
* input size
input size from config
* fix issues with <1.0 erosion
* Update convert.py
* Update Convert_Adjust.py
more work on the way to merginf
* Clean up help note on sharpen
* cleanup seamless
* Delete Convert_Adjust.py
* Update umeyama.py
* Update training_data.py
* swapping
* segmentation stub
* changes to convert.str
* Update masked.py
* Backwards compatibility fix for models
Get converter running
* Convert:
Move masks to class.
bugfix blur_size
some linting
* mask fix
* convert fixes
- missing facehull_rect re-added
- coverage to %
- corrected coverage logic
- cleanup of gui option ordering
* Update cli.py
* default for blur
* Update masked.py
* added preliminary low_mem version of OriginalHighRes model plugin
* Code cleanup, minor fixes
* Update masked.py
* Update masked.py
* Add dfl mask to convert
* histogram fix & seamless location
* update
* revert
* bugfix: Load actual configuration in gui
* Standardize nn_blocks
* Update cli.py
* Minor code amends
* Fix Original HiRes model
* Add masks to preview output for mask trainers
refactor trainer.__base.py
* Masked trainers converter support
* convert bugfix
* Bugfix: Converter for masked (dfl/dfaker) trainers
* Additional Losses (#592)
* initial upload
* Delete blur.py
* default initializer = He instead of Glorot (#588)
* Allow kernel_initializer to be overridable
* Add ICNR Initializer option for upscale on all models.
* Hopefully fixes RSoDs with original-highres model plugin
* remove debug line
* Original-HighRes model plugin Red Screen of Death fix, take #2
* Move global options to _base. Rename Villain model
* clipnorm and res block biases
* scale the end of res block
* res block
* dfaker pre-activation res
* OHRES pre-activation
* villain pre-activation
* tabs/space in nn_blocks
* fix for histogram with mask all set to zero
* fix to prevent two networks with same name
* GUI: Wider tooltips. Improve TQDM capture
* Fix regex bug
* Convert padding=48 to ratio of image size
* Add size option to alignments tool extract
* Pass through training image size to convert from model
* Convert: Pull training coverage from model
* convert: coverage, blur and erode to percent
* simplify matrix scaling
* ordering of sliders in train
* Add matrix scaling to utils. Use interpolation in lib.aligner transform
* masked.py Import get_matrix_scaling from utils
* fix circular import
* Update masked.py
* quick fix for matrix scaling
* testing thus for now
* tqdm regex capture bugfix
* Minor ammends
* blur size cleanup
* Remove coverage option from convert (Now cascades from model)
* Implement convert for all model types
* Add mask option and coverage option to all existing models
* bugfix for model loading on convert
* debug print removal
* Bugfix for masks in dfl_h128 and iae
* Update preview display. Add preview scaling to cli
* mask notes
* Delete training_data_v2.py
errant file
* training data variables
* Fix timelapse function
* Add new config items to state file for legacy purposes
* Slight GUI tweak
* Raise exception if problem with loaded model
* Add Tensorboard support (Logs stored in model directory)
* ICNR fix
* loss bugfix
* convert bugfix
* Move ini files to config folder. Make TensorBoard optional
* Fix training data for unbalanced inputs/outputs
* Fix config "none" test
* Keep helptext in .ini files when saving config from GUI
* Remove frame_dims from alignments
* Add no-flip and warp-to-landmarks cli options
* Revert OHR to RC4_fix version
* Fix lowmem mode on OHR model
* padding to variable
* Save models in parallel threads
* Speed-up of res_block stability
* Automated Reflection Padding
* Reflect Padding as a training option
Includes auto-calculation of proper padding shapes, input_shapes, output_shapes
Flag included in config now
* rest of reflect padding
* Move TB logging to cli. Session info to state file
* Add session iterations to state file
* Add recent files to menu. GUI code tidy up
* [GUI] Fix recent file list update issue
* Add correct loss names to TensorBoard logs
* Update live graph to use TensorBoard and remove animation
* Fix analysis tab. GUI optimizations
* Analysis Graph popup to Tensorboard Logs
* [GUI] Bug fix for graphing for models with hypens in name
* [GUI] Correctly split loss to tabs during training
* [GUI] Add loss type selection to analysis graph
* Fix store command name in recent files. Switch to correct tab on open
* [GUI] Disable training graph when 'no-logs' is selected
* Fix graphing race condition
* rename original_hires model to unbalanced
-scale in detector has been used as a member variable and shared between multiple threads. however scale heavily depends on the concrete image. if thread change occurs between calculation of scale, actual detection and postprocessing, then random behaviour and exceptions might happen. especially in the case of input images with varying sizes this might cause negative effects
-wrong scale factor calculation:
scale factor was calculated as: scale = target / source
whereas target and source are variables for the total number of pixels in the image (= width * height)
However, later on it was applied on the individual dimensions/components as width/height independantly:
e.g.: dims = (int(width * scale), int(height * scale))
or: int(face.left() / scale),
therefore scaling factor often was too small and detection for very big input images was performed on very small intermediate images
Therefore calculation has been changed to:
scale = sqrt(target / source)
torzdf