-Upgrade multiple dependencies: pytorch, cuda, nvidia driver, mmdet
-Additional support for full frame classifiers
-Support viame csv for polygon saveout from desktop annotator
-Seal multi-view GUI fix zooming issue, allow scoring editing
-Fix issue on windows running CFRNN training from pipelines dropdown
-Fix issue in C++ hello world example
-Fix issue in linux running IQR search and rapid model generation
-Fix issue on windows with running CFRNN detectors at test time
-Fix issue with running detector downloading excess un-needed model files
-Fix issue with CFRNN training on windows
-Add continue training from prior model CFRNN script
-Fix issue in habcam model pack
-Fix issue with timeout in latest CFRNN training
-Fix issue with labels.txt file not being used in latest CFRNN training
-Fix scoring examples which became broken in last
-Add tracker training example for windows binaries
-Fix issue in running test-time multi-GPU from an image input list
-Add extra checks for project file scripts with early exits
-Add improved CFRNN training proceedure using netharn
-Fix issue in centos binaries backing X11 library
-Refine cascade faster rcnn training proceedure first pass
-Add tracker-assisted annotation for video
-Fix split track option in default annotator
-New seal multi-view GUI release
-Fix CentOS issues with unable to run pipelines (Ubuntu and Windows fine)
-Updated KWIVER in VIAME causing all add-ons to break and need to be regenerated
-Added in support for object tracking solely based on IOU and image stabilization
-Fixed issue in installers with measurement code after upgrading numpy versions
-Add simple mosaic generation script and related pipelines
-Add multiple seal tracking example pipelines
-Fix issues in prior v0.10.* releases on systems with no CUDA 10s installed
-Remove ball-tree indexing in favor of full LSH hash searches due to issues
-Begin packaging python in binaries for linux in addition to windows
-Baseline speed optimizations and bug fixes for IQR, significant database speedups
-Begin packaging python in binaries for windows, will do this for linux soon
-Fixed windows issues with running and training detectors introduced in last
-Quick fixes to make IQR GUI not crash on images of different size and large images
-Add SVM training tool example, for training SVMs from large amounts of annotations
-Misc tuning and bug fixes
-Lots of bug fixes (windows project files)
-Learning rate tuning for latest detection models
-CUDA 10.0 support
-Add missing .so files into CentOS and Ubuntu binaries to not require external packages
-Support local detection running for multiple variants of motion detectors
-All detector training routines now also generate KWIVER pipelines
-Remove index_existing and merged changes into index_defaults to reduce confusion
-Additional detection models
-New default and generic object detection models
-Improved IQR detector performance
-The default box detector training routine is now cascade rnn instead of yolo
-New detection techniques and default models: cascade faster rcnn, cfrnn with motion, YOLO-WTF
-Additional support for frame-level classification in the system
-Image registration example added, arctic seal examples
-Fixed a few issues in last relating to detector optimizations which broke things in last
-More intuitive final model saveouts
-Windows 7 runtime training tool fixes
-Better automatic video detection, don't process hidden files, dirs recursively or non-videos
-Show correct time offsets in query GUI from video start
-Fix bug added to deep training tool in last version
-Fix training tool windows error reporting
-Fix model saveout in windows binaries broken in last
-Updated default model training parameters
-Minor python change to drop dependency in standard use case
-Add train deep model from GUI dropdown
-Fix issues in prior release dealing with default indexing option and finding local models
-Add YOLOv3, YOLOv3 with Temporal Features detectors
-GUI bug fixes: issues with running embedded pipes and detections not always showing up
-Configuration file cleanup
-Remove multiple models from default installers, move to seperate .zips
-Improved timeline plotting for multi-video inputs
-Fixed issue in image list project files introduced in prior
-More intelligent default track classification probabilities which make use of keyframes
-Add automatic dynamic range compute for 16-bit imagery in GUI
-Expose video rate for image list processing in default project files
-Added extra error checking on indexing scripts involving databases, improved error reporting
-Allow video projects to process directory of directory of images
-Additional error checking on index formation
-Improvements to launch_timeline_viewer script
-Added GUI display dynamic range parameters for 16-bit imagery
-Fix issue with measurement example after config refactor
-Additional bug fixes and parameter tuning relating to tile boundaries when chipping enabled in detectors
-Fixed a bug in debayer and color correction filter .pipe which was causing garbled results in the example
-Added RELEASE_NOTES to binaries
-Fixes to Windows CPU-only binaries
-Fixes for properly processing greyscale imagery
-Non-maximum suppression tuning
-If there is a corrupt frame in the middle of a sequence, keep on processing in both GUI and CLI
-Improved video support in default project file run scripts
-Configuration files have been refactored and their structured cleaned up
-Additional annotation GUI embedded pipelines added, directory structure improved