hi
Enviroment:
- Android 12
- SNPE 1.63.0.3523
- network https://storage.googleapis.com/download.tensorflow.org/models/tflite/pos...
converting to dlc command:
npe-tflite-to-dlc --input_network posenet_mobilenet_v1_100_257x257_multi_kpt_stripped.tflite \
--input_dim serving_default_input:0 "1,257,257,3" \
--output_path posenet_mobilenet_v1_100_257x257_multi_kpt_stripped_new_nod.dlc \
--out_node "MobilenetV1/heatmap_2/BiasAdd" \
--out_node "MobilenetV1/displacement_bwd_2/BiasAdd" \
--out_node "MobilenetV1/displacement_fwd_2/BiasAdd" \
--out_node "MobilenetV1/offset_2/BiasAdd"
Network builder creation code clip:
builder = new SNPE.NeuralNetworkBuilder(application). setRuntimeOrder(runtime). setDebugEnabled(true). setCpuFallbackEnabled(true). setModel(resource, resource.available()). setUseUserSuppliedBuffers(false);
if debug is enabled after processing image it will prodce more otputs than 4 that i put in the conversing command. However if option setDebugEnabled is set to false then we will have only one ouput layer.
Is there any limitation and reason why it is not available more then one output layer if debug is disabled?
Is there any way to overcome this problem?
Best regards
Dear customer,
We don't hear the issue as you descripted. How about snpe-net-run for your converted dlc files?
BR.
Wei
hi Wei,
Execution of the snpe-net-run can be done without problem.
I also can execute the network on the Android with paramiter
setDebugEnabled(true)
we receving all data that is expected from the network.
If the value is set to false
i only get single output layer upon execution of the network and the rest of it not available.
Do you have any SSD metwork with mltiple output nods that i for sure can be used as a base?
Best regards,
Nikola