Run through the example inside ncappzoo/caffe/TinyYolo, take the picture from the camera and test it.
It was found that the processing was slow, and the processing speed was only 2-3 frames per second.
Ncappzoo/tensorflow/tiny_yolo_v2 has the same problem.
What about your test result?
Link Copied
@luna
The size of the graph is too large.
51.68MB
Please reduce it to around 10MB.
https://github.com/PINTO0309/TinyYolo
https://qiita.com/PINTO/items/d5645734ca9c95b1c395
https://qiita.com/PINTO/items/db3ab44a3e2bcd87f2d8
@luna You can use the benchmark ncs app in the ncappzoo to benchmark how long it takes to send and receive an inference on the NCS device. Alternatively, you can view how long it takes for each layer of a model to finish computing using the mvNCProfile tool (see below).
Tiny Yolo V1/V2 both require some post processing that is done on the CPU. Tiny Yolo V1 and V2 both have larger input sizes than most of the other models in the ncappzoo and this could be the reason why they take a little longer to compute the inference result.
$mvNCProfile yolov2-tiny-voc.pb -s 12
...
Time to Execute : 213.38 ms
USB: Myriad Execution Finished
Time to Execute : 187.16 ms
USB: Myriad Execution Finished
USB: Myriad Connection Closing.
USB: Myriad Connection Closed.
Network Summary
Detailed Per Layer Profile
Bandwidth time
# Name MFLOPs (MB/s) (ms)
===============================================================================
0 1-leaky 149.5 537.2 16.592
1 2-maxpool 2.8 792.0 6.668
2 4-leaky 398.7 860.7 13.817
3 5-maxpool 1.4 926.4 2.851
4 7-leaky 398.7 648.6 9.216
5 8-maxpool 0.7 957.0 1.380
6 10-leaky 398.7 403.5 7.716
7 11-maxpool 0.3 948.3 0.696
8 13-leaky 398.7 210.5 9.749
9 14-maxpool 0.2 925.6 0.357
10 16-leaky 398.7 326.9 9.181
11 17-maxpool 0.3 453.1 0.366
12 19-leaky 1594.9 315.1 33.329
13 21-leaky 3189.8 312.5 67.158
14 BiasAdd_8 43.3 415.6 1.382
-------------------------------------------------------------------------------
Total inference time 180.46
-------------------------------------------------------------------------------
Generating Profile Report 'output_report.html'...
@PINTO ,thanks,i will try later!
@Tome_at_Intel thanks, I'll take a look later
For more complete information about compiler optimizations, see our Optimization Notice.