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.
- Currently only
.tflitemodels are supported, and inferencing can be done only on the GPU.
- DSP support will be shipped in a future software update.
When training models to be run on Flo Edge, be sure to check out the operations supported by the GPU
The following code snippets should help you understand using the SDK to inference models.
[inputs] --> model.tflite --> [outputs]
- 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")
- Set inputs
tflite_interface.set_input([input1, input2, ...])
- Invoke model
- Get output
output1, output2, ... = tflite_interface.get_output()
Be sure to check out our examples repo.We’ve added example scripts to run well known models like YOLO v5, MiDAS, etc.