search

PINTO0309 / MobileNet-SSD

MobileNet-SSD(MobileNetSSD) + Neural Compute Stick(NCS) Faster than YoloV2 + Explosion speed by RaspberryPi · Multiple moving object detection with high accuracy.
https://qiita.com/PINTO/
MIT License
92 stars 17 forks source link

Latest SDK (2.08) support #5

Closed yinguobing closed 6 years ago

yinguobing commented 6 years ago

Hi Pinto,

Great work! Looks like Movidius has released a latest NVSDK v2.08.01. I found this version of SDK does not support the graph file included in your repo, Error messages:

W: [         0] checkGraphMonitorResponse:931   Graph monitor request returned error
W: [         0] ncGraphAllocate:1125    The device didn't accept the graph

W: [         0] ncGraphAllocate:1128    graph file version is incompatible

Traceback (most recent call last):
  File "MultiStickSSD.py", line 44, in <module>
    graphHandle.append(graph[devnum].allocate_with_fifos(devHandle[devnum], graph_buffer))
  File "/home/pi/.local/lib/python3.5/site-packages/mvnc/mvncapi.py", line 613, in allocate_with_fifos
    raise Exception(Status(status))
Exception: Status.UNSUPPORTED_GRAPH_FILE

I tried using the graph file compiled localy from /home/pi/Desktop/ncappzoo/caffe/SSD_MobileNet, however, the detection result is always 1 result with class name tvmonitor and a bounding box of whole frame size.

As you have successfully run the SSD model before, would you please try the latest NVSDK and update this repo if possible?

Thanks!

PINTO0309 commented 6 years ago

@yinguobing I did not do anything special, but regenerated the graph file and committed it. It compiled with NCSDK v2.08.01. In addition to "tvmonitor" it seems that "person" also recognizes normally.

By the way, the following repositories have better performance. https://github.com/PINTO0309/MobileNet-SSD-RealSense.git

yinguobing commented 6 years ago

Great! The new graph file works. Also I will try the other repository later. Thanks!