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

How to use dynamic_input function

Model Source

The model used in this example come from the following open source projects:
https://github.com/shicai/MobileNet-Caffe

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 'dynamic_input' parameter of 'rknn.config' is set to:

  dynamic_input = [
      [[1,3,192,192]],    # set 0: [input0_192]
      [[1,3,256,256]],    # set 1: [input0_256]
      [[1,3,160,160]],    # set 2: [input0_160]
      [[1,3,224,224]],    # set 3: [input0_224]
  ]
  

  to simulate models with dynamic input shapes.

Expected Results

This example will print the TOP5 labels and corresponding scores of the test image classification results for each different input shape, as follows:

-----TOP 5-----
Analysing : 100%|███████████████████████████████████████████████| 104/104 [00:00<00:00, 7669.31it/s]
Preparing : 100%|███████████████████████████████████████████████| 104/104 [00:00<00:00, 1079.43it/s]
mobilenet_v1
-----TOP 5-----
[155]: 0.994140625
[154]: 0.0022792816162109375
[204]: 0.001964569091796875
[283]: 0.0009207725524902344
[194 196 284]: 9.40561294555664e-05

Analysing : 100%|███████████████████████████████████████████████| 104/104 [00:00<00:00, 6171.67it/s]
Preparing : 100%|███████████████████████████████████████████████| 104/104 [00:00<00:00, 1598.60it/s]
mobilenet_v1
-----TOP 5-----
[155]: 0.9580078125
[154]: 0.0338134765625
[204]: 0.0074310302734375
[194]: 0.0003018379211425781
[219]: 0.00014257431030273438

Analysing : 100%|███████████████████████████████████████████████| 104/104 [00:00<00:00, 9171.15it/s]
Preparing : 100%|███████████████████████████████████████████████| 104/104 [00:00<00:00, 1389.63it/s]
mobilenet_v1
-----TOP 5-----
[155]: 0.9833984375
[154]: 0.007686614990234375
[204]: 0.0036029815673828125
[283]: 0.0009179115295410156
[193]: 0.0007853507995605469

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