BUG FIXES:

• Fixes bug that causes multiprocessing task to not finish under certain special circumstances.
• Fixes bug that causes certain flows to allocate memory in the GPU even if cpu was selected for usage.

NEW FEATURES:

• Added a simple webhook consumer.

Bug Fixes:

• Fixes bug that can cause flow to not terminate when using multi-task nodes.

Improvements:

• Fixes errors in documentation.
• Removes dependency from scikit-learn

Bug fixes

1. Fixes bug when doing multiprocessing in gpu allocated processors.

New Features

• Detection of bottlenecks
• Improved logging of bottlenecks

Pose and multipose models

classification models

Models released:

• maskrcnn-resnet101_coco, 470 ms per image, 33 COCO mAP
• maskrcnn-inceptionv2_coco, 79 ms per image, 25 COCO mAP

Bug Fixes

• Added missing files to the detection release, and updated source code accordingly.

New Functionality

• Added producers: VideoUrlReader and VideoDeviceReader.

NEW FUNCTIONALITIES:

• Ability to allocate a node to multiple processes
• Flow now stops in a natural way when stopped.

BREAKING CHANGES:

• videoflow.producers.VideoFileReader now returns a tuple of (idx, frame). Before it only returned the frame.

Improvements

• Changed default device_type from 'cpu' to 'gpu' in TensorflowObjectDetector.