JeVois 1.10 brings the following new features:

Improved JeVois Inventor allows for testing of headless (no USB video output) operation of JeVois
Bumped to OpenCV 4.0.0-beta, with all contrib modules including deep neural networks (DNN) and non-free
YOLO v3 is now supported by DarknetYOLO, faster, more accurate, less memory, and 80 object classes.
New module PyClassificationDNN show how to use object recognition deep neural networks using the OpenCV DNN framework, in Python. Supported network formats are Caffe, TensorFlow, Darknet, Torch, and ONNX. Examples are provided which include:
- SqueezeNet v1.1 trained on ImageNet (1000 object categories), Caffe model
- More soon! Please contribute!
Several new tutorials about optical flow implementation in python, headless operation, etc.
Added support for different camera sensors. ov7725 is working great, still work in progress for ov2640 and AR0135 (global shutter, M12 lens mount) with ICM-20948 9-DOF IMU placed right behind the image sensor. More on that coming in the next release.
Added support for on-the-fly conversion from raw BAYER camera frames to YUYV. Useful for sensors which can only output BAYER, like the AR0135 global shutter sensor.
Bugfix of jevois-usbsd which was not working for some users.
Better handling of module startup errors, config file errors, etc so that the Inventor does not get stuck even if a viion module completely fails to load or some config file is bogus.
Additional support for accessing raw camera sensor registers and raw IMU registers, both from C++ and Python.
More docs about optional lenses, optional sensors, etc.
Patch for missing CPU_SETSIZE definition on some machines (needed to compile TensorFlow in jevoisbase).