- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi,
I working on a project to stream video with real time object detection on a web server inside a RPi. I've had some good success with the caffe implementation of mobilenet which you can see here https://github.com/HanYangZhao/NCS-Pi-Stream. I've since tried with the tensorflow version of mobilnets, but it does not seem to work.
What i've done so far:
- Generate the NCS graph using
- Change the image dimension inside my code to 224 x 224 pixels\
- Change the label to refer to labels.txt
It seems that the object returned by the handle.GetResults() function is different than the one in Caffe implementation(different length) Am I correct? Is there a guide on how to process this object?
Thanks
- Tags:
- Tensorflow
Link Copied
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
@bobthesheep I could be wrong, but I think you may have mixed up the networks (object detectors and image classifiers).
When you are saying that the tensor "returned by the handle.GetResults() function is different than the one in Caffe implementation", are you referring to this network mobilenets?
The network being used in the github repository you linked (https://github.com/HanYangZhao/NCS-Pi-Stream) is SSD MobileNet and the network you may be referring to is MobileNet (TensorFlow).
The difference is that MobileNet is an image classifier and SSD MobileNet is an object detector.
With respect to SSD MobileNet for TensorFlow, the current NCSDK (1.12) does not have support for this network yet.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Ohhh I see my bad. Thanks for clarifying.
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page