rknn-toolkit2/examples/functions/dynamic_shape/README.md

How to use dynamic shape 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 = [
      [[1,3,256,256]],    # 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:

--> Running model with input shape [1,3,224,224]
GraphPreparing : 100%|██████████████████████████████████████████| 104/104 [00:00<00:00, 4764.33it/s]
SessionPreparing : 100%|█████████████████████████████████████████| 104/104 [00:00<00:00, 366.73it/s]
-----TOP 5-----
[155] score:0.993652 class:"Shih-Tzu"
[154] score:0.002277 class:"Pekinese, Pekingese, Peke"
[204] score:0.002277 class:"Lhasa, Lhasa apso"
[283] score:0.000673 class:"Persian cat"
[196] score:0.000109 class:"miniature schnauzer"
--> Running model with input shape [1,3,160,160]
GraphPreparing : 100%|██████████████████████████████████████████| 104/104 [00:00<00:00, 9800.88it/s]
SessionPreparing : 100%|████████████████████████████████████████| 104/104 [00:00<00:00, 1296.61it/s]
-----TOP 5-----
[155] score:0.963867 class:"Shih-Tzu"
[154] score:0.029099 class:"Pekinese, Pekingese, Peke"
[204] score:0.006393 class:"Lhasa, Lhasa apso"
[194] score:0.000415 class:"Dandie Dinmont, Dandie Dinmont terrier"
[219] score:0.000090 class:"cocker spaniel, English cocker spaniel, cocker"
--> Running model with input shape [1,3,256,256]
GraphPreparing : 100%|██████████████████████████████████████████| 104/104 [00:00<00:00, 8788.48it/s]
SessionPreparing : 100%|████████████████████████████████████████| 104/104 [00:00<00:00, 1404.91it/s]
-----TOP 5-----
[155] score:0.981445 class:"Shih-Tzu"
[154] score:0.008896 class:"Pekinese, Pekingese, Peke"
[204] score:0.004169 class:"Lhasa, Lhasa apso"
[193] score:0.000783 class:"Australian terrier"
[283] score:0.000783 class:"Persian cat"

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