- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
I have generated code using model optimizer using following command:
./ModelOptimizer -w ~/caffe.model -p FP16 -d ~/tiny.prototxt --target APLK --network CLASSIFICATION -f 1.0 -dm -c
I have compiled the code which generated using above command.
I have compiled the code the generated. I am running the binary as follows:
./generated_code
I am getting following errors:
.............................
.............................
- fc12_weights.bin
- fc12_bias.bin
VXLOG: Node <node>[0x185c0b0] (org.khronos.nn_extension.convolution_layer) parameter[0] failed validation
(sts=VX_ERROR_NOT_SUPPORTED)
ERROR: failed to verify graph (vx_status=VX_ERROR_NOT_SUPPORTED)
May I know the reason for this ?
Link Copied
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi Deepak,
From the error, it seems that there is a validation issue with parameter 0. the parameters the graph is expecting, are defined in graphprocess.c file. since you did not pass on any parameters in your ./generated_code command, i assume that is the reason for the error.
You can look at the graphprocess.c file yourself, or if you need help you can attach it to this thread and i will try to help you with it.
Regards,
Stav
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
I have generated optimized model files and codes using ModelOptimizer by specifying batch size =1 as command line argument in ModelOptimizer (In my prototxt file the batch size is 1). Now I am able to run the generated code succesfully.
(I think in ModelOptimizer default batch size is 4, the error was due to the mismatch in batch size specified in the prototxt and ModelOptimizer)
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi Deepak,
the default is set by the number of cores you have in your system. Happy to hear it's working.
Regards,
Stav
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page