ontheway16

Hello, yes it doesn't work with "width" etc. You have to use xmin, ymin and xmax, ymax. Means; x, y, x+width, y+height, as you stated above.

You may experiment with the following, https://github.com/NVIDIA/caffe/blob/caffe-0.15/examples/kitti/detectnet_network-2classes.prototxt But, for multiple classes, things are not so easy with detectnet, if you have more than 10-20 classes. Even with 2-3 classes, I...

Please check the following; https://github.com/NVIDIA/caffe/pull/157 https://github.com/NVIDIA/DIGITS/tree/digits-4.0/digits/extensions/data/objectDetection#custom-class-mappings

I made a diff and found the following: ``` layer { name: "cvg/classifier" type: "Convolution" bottom: "pool5/drop_s1" top: "cvg/classifier" param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2...

As far as I know, it trains for both at the same time. But be prepared to see lower mAP levels than single class trainings. Thats why I said, first...

> It looks like you have to change the layer name as you are training a custom data set. I had the same problem and changing the name of few...

> > I am planning to invest on Olympus M5 mark3, solely to use with libgphoto2, from a RasPi device. > > Make sure there is a return policy that...

Apparently, m5 mk3 has no USB tethering support, which answers my above questions.

Very unfortunate for mark III, but it seems they pruned that functionality, by hardware. I will switch my route either to Oly M1 m2, or Pana G9. Currently i have...

> It also **seems** that the new OM System OM-5 (successor to the E-M5 Mark III) supports tethering. But then it will have to be added to libgphoto2. I hope...