- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
I fine tuned the VGG16 face model with 6 output class with my own data, when testing I found it takes 0.78s to finish the following code:
@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
graph.LoadTensor(img.astype(numpy.float16), 'user object')
output, userobj = graph.GetResult()
@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
the img is 3x224x224 picture, and surely it comes the right output, but I am wondering whether the 0.78s is right, I think it's a bit long.
Link Copied
2 Replies
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
@fengping It could be due to a number of things. I would first recompile the graph file and make sure that you are using the -s 12 option. Also you can use the mvNCProfile tool to time the different layers in your network. It provides the amount of time taken to process the layer, the bandwidth the layer uses in MB/sec and MFLOPs. mvNCProfile also provides a graphical representation of your network with this information. You can find out more by visiting https://movidius.github.io/ncsdk/tools/profile.html.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
@Tome_at_Intel
I do used the -s 12 option when compiling. I saw the log when sending data to ncs: Successfully sent 865212 bytes of data in 82.947172 ms (9.947662 MB/s)
And I try to using the mvNCProfile to have the information of each layer, but got an error which said "[Error 13] Toolkit Error: Provided OutputNode/InputNode name does not exist or does not match with one contained in model file Provided: VGGNet.graph/conv1_1".
I am sure that the conv1_1 and prob layer exist in the deploy.prototxt both usingmvNCCompile and mvNCProfile.
Reply
Topic Options
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page