Running deeplearning models

Flo Edge has out of the box capabilities for Deep learning. It comes equipped with Qualcomm’s Adreno 630 GPU and Hexagon 685 DSP along with an SDK to harness their compute.

Important

Inferencing

• Currently only .tflite models are supported, and inferencing can be done only on the GPU.
• DSP support will be shipped in a future software update.

Training

When training models to be run on Flo Edge, be sure to check out the operations supported by the GPU

Using the SDK

The following code snippets should help you understand using the SDK to inference models.

[inputs] --> model.tflite --> [outputs]

1. Load model

     from anx_interface import TfliteInterface, DeviceType
     tflite_interface = TfliteInterface(DeviceType.GPU)
     # success is True if model loaded correctly else False
     success = tflite_interface.load_model("/path/to/model.tflite")
   

2. Set inputs

     tflite_interface.set_input([input1, input2, ...])
   

3. Invoke model

     tflite_interface.invoke()
   

4. Get output

     output1, output2, ... = tflite_interface.get_output()
   

Examples

Be sure to check out our examples repo.We’ve added example scripts to run well known models like YOLO v5, MiDAS, etc.