- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
As we known, MTCNN has multi model and multi output layer. I want to use one NCS1 load three model. First I find a MTCNN repo https://github.com/happynear/MTCNN_face_detection_alignment/tree/master/code/codes/MTCNNv2/model , then I use mvNCCheck det1.proto and det1.caffemodel(proto change input_shape),
I get a error :
#
E: [0] dispatcherEventReceive:200 dispatcherEventReceive() Read failed -4
E: [0] releasePacketFromStream:362 There is no packet to release
E: [0] dispatcherEventSend:730 Write failed -4
E: [0] dispatcherEventSend:730 Write failed -4
Traceback (most recent call last):
File "/usr/local/bin/mvNCCheck", line 202, in
quit_code = check_net(args.network, args.image, args.inputnode, args.outputnode, args.nshaves, args.inputsize, args.weights, args)
File "/usr/local/bin/mvNCCheck", line 176, in check_net
timings, myriad_output = run_myriad(graph_file, args)
File "/usr/local/bin/ncsdk/Controllers/MiscIO.py", line 261, in run_myriad
graph.destroy()
File "/usr/local/lib/python3.5/dist-packages/mvnc/mvncapi.py", line 775, in destroy
status = f.ncGraphDestroy(byref(self.handle))
#
And I find @Tome_at_Intel said NCS only support one output layer https://ncsforum.movidius.com/discussion/1084/question-on-multi-output-model-does-ncsdk-support-multi-output, So I want to use concat(caffe) solves this problem.
I find this https://ncsforum.movidius.com/discussion/comment/2831/#Comment_2831, I used reshape layer but it dose't work. I get the same error.
Last, I find NCSDK v2.08 can't loaded more than two graph, but NCSDK v2.05 can do it. ???
Thanks for any reply or discussion!!
- Tags:
- Caffe
Link Copied
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
@curry_best I can confirm that you're correct about the following:
- NCS only supports one output.
- NCSDK 2.08 currently has a bug where it cannot load more than 2 graphs on one device.
- Currently MTCNN is not working with the current NCSDK (2.08)
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
@Tome_at_Intel thanks Tome, Can one NCS2 with OPENVINO implement MTCNN without problem? And why SSD_MobileNet can use concat layer?
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
@curry_best Regarding MTCNN with NCS2 and OpenVINO, that would be a better question to be posted on the OpenVINO forums.
For concat and SSD MobileNet on Caffe, you should be able to concat multiple outputs together if you want but you have to use a dummy reshape layer as the final layer after your concat layer. There is currently an issue where the concat layer is supported on the NCSDK when it is the last layer. This post may help: https://ncsforum.movidius.com/discussion/comment/3503/#Comment_3503. Thanks.
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page