为什么跑mnist例程的代码，正确率这么低

[cgti]

请在这里粘贴代码

# This code run in OpenMV4 H7 or OpenMV4 H7 Plus

import sensor, image, time, os, tf

sensor.reset()                         # Reset and initialize the sensor.
sensor.set_pixformat(sensor.GRAYSCALE)    # Set pixel format to RGB565 (or GRAYSCALE)
sensor.set_framesize(sensor.QVGA)      # Set frame size to QVGA (320x240)
sensor.set_windowing((240, 240))       # Set 240x240 window.
sensor.skip_frames(time=2000)          # Let the camera adjust.

clock = time.clock()
while(True):
    clock.tick()
    img = sensor.snapshot().binary([(0,64)])
    for obj in tf.classify("trained.tflite", img, min_scale=1.0, scale_mul=0.5, x_overlap=0.0, y_overlap=0.0):
        output = obj.output()
        number = output.index(max(output))
        print(number)
    print(clock.fps(), "fps")

[uyvo]

我也这么觉得蹲个解决方法

[gwum]

mnist为什么直接运行例程可以，设置roi区域后根本就不能识别roi区域里的数字？

[gwum]

@gwum 而且必须离得很近，根本就没有使用的场景，我不知道快贴到数字才能识别有什么意义？

[kidswong999]

@gwum https://github.com/SingTown/openmv_tensorflow_training_scripts/blob/main/mnist/openmv_mnist.ipynb

代码在这里，你可以自己改数据集，想怎么改就怎么改。