【周易AIPU 仿真】使用周易 AIPU 部署mobilenet_v1及仿真

1. 原模型模型下载模型
   百度云盘下载: 链接 提取码: kpuh
2. 测试图片Shetland_sheepdog:

3.矫正集

将图片和label放到calibration目录下的img路径和label.txt下，
执行 preprocess_for_dataset.py 生成 dataset.npy和label.npy矫正集文件。

4.cfg

如附件中 tflite_mobilenet_v1_run.cfg 所示：

[Common]
mode=run

[Parser]
model_type = tflite
model_name = mobilenet_v1  
detection_postprocess = 
model_domain = image_classification
output = MobilenetV1/Predictions/Reshape_1
input_model = ./mobilenet_v1_1.0_224.tflite
input = input
input_shape = [1,224,224,3]
output_dir = ./

[AutoQuantizationTool]
model_name = mobilenet_v1
quantize_method = SYMMETRIC
ops_per_channel = DepthwiseConv
calibration_data = ./dataset/dataset.npy
calibration_label = ./dataset/label.npy
preprocess_mode = normalize
quant_precision=int8
reverse_rgb = False
label_id_offset = 0

[GBuilder]
inputs=./model/input.bin
simulator=aipu_simulator_z1
outputs=output_mobilenet.bin
profile= True
target=Z1_0701

5.simulator 执行结果
aipubuild config/tflite_mobilenet_v1_run.cfg
运行之后得到 output_mobilenet.bin，使用 quant_predict.py 解析后：

predict first 5 label:
    index  231, prob 192, name: Shetland sheepdog, Shetland sheep dog, Shetland
    index  232, prob  47, name: collie
    index  260, prob   6, name: Pomeranian
    index  262, prob   4, name: keeshond
    index  233, prob   3, name: Border collie

从结果中可以看出预测到Shetland sheepdog(设得兰牧羊犬).

相关文件请见附件。