android13/hardware/rockchip/rknn-toolkit2/examples/functions/multi_input_test

How to convert multi-input model

Model Source

The model used in this example come from 'gen_pb.py'.

Script Usage

Usage:

python test.py

Description:

• The default target platform in script is 'rk3566', please modify the 'target_platform' parameter of 'rknn.config' according to the actual platform.
• If connecting board is required, please add the 'target' parameter in 'rknn.init_runtime'.
• The model has multiple inputs, so multiple input files need to be set in 'dataset.txt':

  dog_128x128.jpg input2.npy input3.npy dog_128x128_gray.png
  

• The 'inputs' and 'data_format' of 'rknn.inference' should also be set accordingly, as follows:

  rknn.inference(inputs=[img, input2, input3, img_gray], data_format=['nhwc', 'nchw', 'nchw', 'nhwc'])
  

Expected Results

This example will print outputs of inference, as follows:

inference result:  [array([[ -3.1041162,  -7.0548096,  -5.361655 , ..., -14.391811 ,
        -15.802773 , -11.287695 ]], dtype=float32)]

• Note: Different platforms, different versions of tools and drivers may have slightly different results.