基于Aidlux平台的工业视觉少样本缺陷检测

39 阅读1分钟

工业视觉缺陷检测的工作流程

image.png 常用异常检测算法 image.png

面临的挑战及发展

image.png

image.png 图像分割的数据标注

image.png

image.png 数据标注准确的重要性:

1. 训练模型的基础

2. 提高模型性能

3. 降低误判和误诊分险

4. 减少资源浪费

自动标注SAM的使用

image.png 模型切换

image.png 模型部署

# -*- coding: UTF-8 -*-

import aidlite_gpu

import cv2

import os

import time

import numpy as np

from PIL import Image

 

import matplotlib.pyplot as plt

def mask_to_image(mask: np.ndarray):

    if mask.ndim == 2:

        return Image.fromarray((mask * 255).astype(np.uint8))

    elif mask.ndim == 3:

        return Image.fromarray((np.argmax(mask, axis=0) * 255 / mask.shape[0]).astype(np.uint8))

  


def aidlux_tflite_infer(model_path, img_path, save_path):

    # step1: 初始化aidlite类并创建aidlite对象

    aidlite = aidlite_gpu.aidlite()

    print('model initial success!!')

 

    # step2: 加载模型

    inp_shape = [256*256*1*4]

    out_shape = [256*256*2*4]

    value = aidlite.ANNModel(model_path, inp_shape, out_shape, 4, 0)

    # step3: 传入模型输入数据

    img = cv2.imread(img_path, 0)

    img = cv2.resize(img, (256, 256))

    img = img[np.newaxis, ...]

    img = img / 255.0

    img = np.expand_dims(img, axis=0)

    img = img.astype(dtype=np.float32)

    print("image shape is ", img.shape)

    aidlite.setInput_Float32(img)

   

    # step4: 执行推理

    start = time.time()

    aidlite.invoke()

    end = time.time()

    print("infer time(ms):{0}", 1000 * (end - start))

    # step5: 获取输出

    pred = aidlite.getOutput_Float32(0)

    # step6: 后处理

    pred = np.array(pred)

    pred = np.reshape(pred,(2,256,256))

    mask_img = mask_to_image(pred)

   

    mask_img.save(save_path)

    # mask_img = np.array(mask_img)  

    # cv2.imshow('mask_img', mask_img)

    # cv2.waitKey(0)

    # cv2.destroyAllWindows()

   

if __name__ == '__main__':

    model_path = "/home/dataset2aidlux/unetmodel_fp32.tflite"

    img_path = "/home/dataset2aidlux/test_imgs/0597.PNG"

    save_path = '/home/dataset2aidlux/test_imgs/result_0597.png'

    aidlux_tflite_infer(model_path, img_path, save_path)

效果视频: 基于Aidlux的语义分割模型转换:www.bilibili.com/video/BV1K6… 基于Aidlux的语义分割模型部署:www.bilibili.com/video/BV19u…