It looks like you're new here. If you want to get involved, click one of these buttons!
Please feel free to add comments if you choose "other" for any of the following questions.
An IMPORTANT user feedback is better documentation. The dev team please put sometime for that stuff. At the moment, NCS is great but poorly documented, and the Docs website is not properly constructed, like information is scattered and put anywhere possible. A neat and concise doc like those of Tensorflow or Keras or MxNet is greatly appreciated. Thanks!
A user in ROS Movidius NCS ask for YOLO and FasterRCNN support(https://github.com/intel/ros_intel_movidius_ncs/issues/87)
Seeing a lot of users having USB integration problems with Movidius, will Intel consider releasing a firmware patch to allow user to choose to default to USB 3.0 permanently? Issues will surface based on current architecture where Movidius will default to USB 2 and then to USB 3 when there is a need to transfer huge data and then back to USB 2 again.
Another feedback is to have more platforms for the community to give feedback on the tools. Right now, the "Issues" tab on the Github repo is disabled. It would be extremely easy to get community feedback on the repo because it's visited by everyone who will get started with the stick. Moreover, people can search for similar issues raised by other users and even possible solutions. Anything more complicated could be taken in the forum.
Support of ONNX in the future would be nice.
What use cases are you most interested in? I think this question falls into what is the common
terminology of the day issue.
Yolo segments the image and classifys the objects in the image, that checks multiple boxes as does
what Tensorflow calls this Tensorflow Object Detection API
Is that what you are asking when you say Object detection? But isn't Tensorflows Object detection break down into segmenting
then classifying the image segments of image giving the same result as YOLO?
Or what do you call classifications?
is there any plan to support more types of layer?
@dvpo Thanks for the feedback. We really appreciate your input!
@xhuan28 Yes we are planning to support more layers and features in future releases of the NCSDK.
@Tome_at_Intel , thanks for your prompt reply. That would be great you are planning to support more types of layer. From my perspective, I want ROIAlign and ROIPooling which are used in faster RCNN and FPN to be supported in future.
Primary current network interest is SSD-MobleNet. I would especially like a tensorflow version to run on the NCS. In the near future my interest will probably include the mozilla Deep Speech model and face recognition.
Tensor Flow Object Detection API