Recent Releases of openpifpaf
openpifpaf - v0.13.10
- refactored loss components, no change in output
- Python
Published by svenkreiss over 3 years ago
openpifpaf - v0.13.9
- minor fixes
- remove double soft-clamp applied to Scale losses, now this clamp is applied within the components
- Python
Published by svenkreiss over 3 years ago
openpifpaf - v0.13.8
- PyTorch 1.13.1 and TorchVision 0.14.1
- configurable evaluator
- train --out accepts a directory and auto-adds filename (#587)
- Python
Published by svenkreiss over 3 years ago
openpifpaf - v0.13.7
- PyTorch 1.13.0 and TorchVision 0.14.0
- Python
Published by svenkreiss over 3 years ago
openpifpaf - v0.13.6
- compatibility fixes (Pillow, pycocotools)
- CI: new scheduled tests
- Python
Published by svenkreiss over 3 years ago
openpifpaf - v0.13.5
- PyTorch 1.12.1 and TorchVision 0.13.1
- factor out openpifpaf_extras
- versioneer 0.21 -> 0.23
- Python
Published by svenkreiss almost 4 years ago
openpifpaf - v0.13.4
- restrict protobuf version
- compatibility for future PIL versions
- patch for pretrained MobileNetv3 models
- Python
Published by svenkreiss about 4 years ago
openpifpaf - v0.13.3
- fixes for Python 3.10
- use relative paths for cpp_extension source files
- Python
Published by svenkreiss about 4 years ago
openpifpaf - v0.13.2
- add zip source distribution to PyPi for Windows
- upgrade to torch==1.11.0 and torchvision==0.12.0
- add "suggest edit" button to Guide
- Python
Published by svenkreiss about 4 years ago
openpifpaf - v0.13.1
- fix torch==1.9.0 and torchvision==0.10.0 to solve install issues
- Python
Published by svenkreiss over 4 years ago
openpifpaf - v0.13.0
- major rewrite of the decoders to C++ backed by libtorch
- Python
Published by svenkreiss over 4 years ago
openpifpaf - v0.12.12
- constrain Pillow dependency below 8.3 for torchvision compatibility
- Wholebody fixes #451
- det decoder: nms by category and minor fix to nms suppression
- debug: properly show crowd annotations in encoder debug plots
- Python
Published by svenkreiss almost 5 years ago
openpifpaf - v0.12.11
- wheels for aarch64
- new pretrained models: mobilenetv3large and resnet18-cocodet
- show: default output is now jpeg with image-dpi-factor=2.0
- ONNX export with dynamic batch dimension
- Python
Published by svenkreiss almost 5 years ago
openpifpaf - v0.12.10
- new Guide tutorials: Wholebody, Apollo, Animals
- Predictor API
- MobileNetv3 backbone
- BCE background clamp
- support and CI tests for Python 3.9
- fix for multiple metrics
- Python
Published by svenkreiss about 5 years ago
openpifpaf - v0.12.9
- new tutorials: Custom Dataset and ApolloCar3D
- bugfix: inverse ground-truth annotation for metric #399
- scipy is now an optional dependency
- binaries for Apple Silicon (universal2 and arm64)
- automatically register plugins
- pad before RotateUniform
- area-of-interest based cropping now uses bounding box annotations and not keypoints
- matplotlib 3.4 compatibility
- ONNX 1.9 compatibility
- Python
Published by svenkreiss about 5 years ago
openpifpaf - v0.12.8
- moved to github.com/openpifpaf/openpifpaf
- Python
Published by svenkreiss about 5 years ago
openpifpaf - v0.12.7
- Apollo3D car model with 66 keypoints #376
- new pretrained models for shufflenetv2k16 and shufflenetv2k30
- new transforms.RotateUniform and transforms.RandomChoice #380
- cython decoder.utils.Occupancy #381
- refactor Stream source for video data #371
- additional tests and minor changes for jit.script compatibility #377
- improve visualization of occupancy map for debugging
- Python
Published by svenkreiss about 5 years ago
openpifpaf - v0.12.6
- new training option:
--clip-grad-value - new argument for video tool:
--video-output=virtualcam - faster training startup when training from checkpoint: #363
- test PyTorch 1.8
- eval tool: show all available eval stats by default
- Python
Published by svenkreiss about 5 years ago
openpifpaf - v0.12.5
- speed improvement for
openpifpaf.video#362 - new paper on arxiv: https://arxiv.org/abs/2103.02440
- Python
Published by svenkreiss about 5 years ago
openpifpaf - v0.12.4
A custom dependency blocked the upload of v0.12.3 to PyPI and this should be fixed now.
Other changes:
* new datasets.NumpyImageList
* OpenCV tutorial in Guide
- Python
Published by svenkreiss over 5 years ago
openpifpaf - v0.12.3
Minor. Mostly updates to the Guide.
- Python
Published by svenkreiss over 5 years ago
openpifpaf - v0.12.2
Minor bugfixes for exports for recent v0.12 release.
- Python
Published by svenkreiss over 5 years ago
openpifpaf - v0.12.1
Minor release on top of v0.12. See previous release notes.
Here:
* fix Guide deploys
* add --no-sync-batchnorm option
- Python
Published by svenkreiss over 5 years ago
openpifpaf - v0.12
Major rewrite with many backwards incompatible changes. The Guide is updated: https://vita-epfl.github.io/openpifpaf/intro.html
Major changes:
* plugins: many components are now customizable with independent plugins; see Guide: Plugins Overview
* Datamodule: a new class that holds all information related to datasets. Create a new Datamodule to add a new dataset to OpenPifPaf. Cifar10 is now implemented as a plugin and documented.
* Metric: now you can implement your own with a plugin.
* train on multiple datasets simultaneously
* export to CoreML for iOS
* export to onnx with full test with onnxruntime
* --show and --save-all to either show or store all debug images as files for remote debugging
* experimental support for DistributedDataParallel (DDP) training
- Python
Published by svenkreiss over 5 years ago
openpifpaf - v0.12b4
Remove Cython from build process for releases.
- Python
Published by svenkreiss over 5 years ago
openpifpaf - v0.12b3
Generalize some interfaces. Softclamp.
- Python
Published by svenkreiss over 5 years ago
openpifpaf - v0.12b2
Focal loss stability improved.
- Python
Published by svenkreiss over 5 years ago
openpifpaf - v0.12a5
Replace Exp() activations with Softplus() activations. This is backwards incompatible: it degrades significantly performance of previously trained models. A new pretrained shufflenetv2k16 model is available.
- Python
Published by svenkreiss over 5 years ago
openpifpaf - v0.12a2
Fix versioneer in wheel and sdist.
- Python
Published by svenkreiss over 5 years ago
openpifpaf - v0.12.alpha1
Testing automatic publishing to pypi and binary wheels.
- Python
Published by svenkreiss over 5 years ago
openpifpaf - v0.11.9
- new
--val-intervalwhich replaces the use-cases for previous--duplicate-data - extended ColorJitter ranges for training: brightness=0.4, saturation=0.4, hue=0.1
- PyLint 2.6 compatibility
- matplotlib 3.3 compatibility
- Cython 0.29.21
- Python
Published by svenkreiss almost 6 years ago
openpifpaf - v0.11.8
- PyTorch 1.6 compatibility
- modified loss recommendation to avoid autotune
- use versioneer to create dynamic version numbers
- use stretch augmentation #262
- improved decoding performance #252
- add v0.8 cpp decoder #249
- Python
Published by svenkreiss almost 6 years ago
openpifpaf - v0.11.7
- fix predict with multiple GPUs
- fix ONNX optimization
- fix dense connection option and add a test for it
- suppress a warning during training that doesn't apply
- Python
Published by svenkreiss almost 6 years ago
openpifpaf - v0.11.6.post1
Fix logo and favicon for guide.
- Python
Published by svenkreiss almost 6 years ago
openpifpaf - v0.11.6
- circular truncation for CifHr
- extend Guide
- Python
Published by svenkreiss almost 6 years ago
openpifpaf - v0.11.5
update README and reference Guide
- Python
Published by svenkreiss almost 6 years ago
openpifpaf - v0.11.4.post1
rebuild docs with fixed binder link
- Python
Published by svenkreiss almost 6 years ago
openpifpaf - v0.11.3
- new debug message for "neural network device" (to check cpu versus gpu usage)
- debug output without plots is default; enable debug plots with new
--debug-images
- Python
Published by svenkreiss about 6 years ago
openpifpaf - v0.11.2
- pretrained resnet50 model
- fix CUDA support in
openpifpaf.video - add
--versionoption to all CLIs
- Python
Published by svenkreiss about 6 years ago
openpifpaf - v0.11.0
- major refactor
- now requires Python>=3.6 for type annotations
- new ShuffleNetV2 models:
shufflenetv2k16wandshufflenetv2k30w - 64bit loss and Focal Loss for confidences
- fast fused convolutions for
CompositeHeadFused - new handling of crowd annotations in encoder
- new
--extended-scaletraining and eval-coco option - decoding with frontier is default
- more robust blending of connection candidates
- introduced
openpifpaf.visualizerand many improvements to visualization - [experimental] new
cocodetdataset interface for detections
- Python
Published by svenkreiss about 6 years ago
openpifpaf - v0.10.1
- faster decoder
- refactored scale generation between loss and encoder
- Python
Published by svenkreiss over 6 years ago
openpifpaf - v0.10.0
- major refactor: move all factory-code into factories
- new experimental decoder
- improved image rescaling randomization
- module-level logging
- index-matching bug fixed by @junedgar #147
- tests for Windows, PyTorch 1.3 and with pylint 2.4
- Python
Published by svenkreiss over 6 years ago
openpifpaf - v0.9.0
- make image transforms part of preprocessing #100
- field-based two-scale implementation #101, also modifies single-scale decoder
- added a
show.CrowdPainterto visualize crowd annotations - new benchmark script and updated performance numbers in README.md, #104
- Python
Published by svenkreiss almost 7 years ago
openpifpaf - v0.8.0
- add support for
resnext50,shufflenetv2x1andshufflenetv2x2 - new pretrained models
- new transforms.RandomApply() and transforms.RotateBy90(); removed old transforms
- new blur augmentation
- improved BCE masks #87
- Python
Published by svenkreiss almost 7 years ago
openpifpaf - v0.7.0
- faster seed generation in decoder
- training log plot improvements (labels, consistent colors)
- improved debug visualizer for decoder
- Python
Published by svenkreiss almost 7 years ago
openpifpaf - v0.6.3
- support parallel decoding for
predictandeval_coco(~4x speed improvement) which is automatically activated for batch sizes larger than 1
- Python
Published by svenkreiss about 7 years ago
openpifpaf - v0.6.2
- improved decoder performance #61, #63, #64
- remove
apply_class_sigmoidproperty from head nets and use the standardmodel.train()andmodel.eval()methods instead - bugfix for runs with padding
- improved log messages
- log the names of the fields (preparation to improve plots of training losses)
- Python
Published by svenkreiss about 7 years ago
openpifpaf - v0.6.0
- Torch 1.1.0 compatibility (older versions work but have modified learning rate schedule due to https://github.com/pytorch/pytorch/pull/7889)
- more aggressive NMS
- multi-scale support for
eval_coco:--two-scale,--three-scale,--multi-scale - guaranteed complete poses with
--force-complete-pose
- Python
Published by svenkreiss about 7 years ago
openpifpaf - v0.5.0
- new pre-trained models
- preprocessing rewrite
- decoder profiling update and speed improvements
- Python
Published by svenkreiss about 7 years ago
openpifpaf - v0.4.0
Plugin architecture for heads, losses, encoders and decoders.
- Python
Published by svenkreiss about 7 years ago
openpifpaf - v0.3.0
- API change: single processor returned from decoder factory
- returned keypoint confidences are now a factor of 4 lower (instance confidences are unaffected)
- Python
Published by svenkreiss about 7 years ago