deeplab - v3解读及语义分割度量方法介绍

原文链接：https://baijiahao.baidu.com/s?id=1595995875370065359&wfr=spider&for=pc

该文章详细介绍了deeplab-v3网络中用到的resnet、空洞卷积、空间空洞金字塔池化等内容，并对部分代码进行了详细解读，理解比较容易。

语义分割模型常用的度量标准/准确度：pixel accuracy, mean accuracy, mean IU, frequency weighted IU

原文链接：https://blog.csdn.net/lilai619/article/details/80065013

关键部分来了，如何变成代码呢，有以下两种方案：

（1）参考这个连接：https://github.com/martinkersner/py_img_seg_eval 【已验证，好用】

（2）使用下方原博文的代码【未验证】

下面是根据全卷积语义分割的准确度程序编写

import _init_paths
import os
import numpy as np
from PIL import Image
import matplotlib.pyplot as plt
from skimage import io
from timer import Timer
import cv2
from datetime import datetime
import caffe

test_file = 'test.txt'
file_path_img = 'JPEGImages'
file_path_label = 'SegmentationClass'
save_path = 'output/results'
test_prototxt = 'Models/test.prototxt'
weight = 'Training/Seg_iter_10000.caffemodel'
layer = 'conv_seg'
save_dir = False # True

if save_dir:
    save_dir = save_path
else:
    save_dir = False

# load net
net = caffe.Net(test_prototxt, weight, caffe.TEST)

# load test.txt
test_img = np.loadtxt(test_file, dtype=str)

def fast_hist(a, b, n):
    k = (a >= 0) & (a < n)
    return np.bincount(n * a[k].astype(int) + b[k], minlength=n**2).reshape(n, n)

# seg test

print('>>>', datetime.now(), 'Begin seg tests')

n_cl = net.blobs[layer].channels

hist = np.zeros((n_cl, n_cl))

# timers

_t = {'im_seg' : Timer()}

# load image and label

i = 0
for img_name in test_img:
    _t['im_seg'].tic()
    img = Image.open(os.path.join(file_path_img, img_name + '.jpg'))
    img = img.resize((512, 384), Image.ANTIALIAS)
    in_ = np.array(img, dtype=np.float32)
    in_ = in_[:,:,::-1] # rgb to bgr
    in_ -= np.array([[[68.2117, 78.2288, 75.4916]]])#数据集平均值，根据需要修改
    in_ = in_.transpose((2,0,1))
    label = Image.open(os.path.join(file_path_label, img_name + '.png'))
    label = label.resize((512, 384), Image.ANTIALIAS)#图像大小（宽，高），根据需要修改
    label = np.array(label, dtype=np.uint8)
    # shape for input (data blob is N x C x H x W), set data
    net.blobs['data'].reshape(1, *in_.shape)
    net.blobs['data'].data[...] = in_
    net.forward()
    _t['im_seg'].toc()
    print('im_seg: {:d}/{:d} {:.3f}s' \
          .format(i + 1, len(test_img), _t['im_seg'].average_time))
    i += 1
    hist += fast_hist(label.flatten(), net.blobs[layer].data[0].argmax(0).flatten(), n_cl)
    if save_dir:
        seg = net.blobs[layer].data[0].argmax(axis=0)
        result = np.array(img, dtype=np.uint8)
        index = np.where(seg == 1)
        for i in range(len(index[0])):
            result[index[0][i], index[1][i], 0] =  255
            result[index[0][i], index[1][i], 1] =  0
            result[index[0][i], index[1][i], 2] =  0
        result = Image.fromarray(result.astype(np.uint8))    
        result.save(os.path.join(save_dir, img_name + '.jpg'))        
iter = len(test_img)
# overall accuracy
acc = np.diag(hist).sum() / hist.sum()
print('>>>', datetime.now(), 'Iteration', iter, 'overall accuracy', acc)
# per-class accuracy
acc = np.diag(hist) / hist.sum(1)
print('>>>', datetime.now(), 'Iteration', iter, 'mean accuracy', np.nanmean(acc))
# per-class IU
iu = np.diag(hist) / (hist.sum(1) + hist.sum(0) - np.diag(hist))
print('>>>', datetime.now(), 'Iteration', iter, 'mean IU', np.nanmean(iu))
freq = hist.sum(1) / hist.sum()
print('>>>', datetime.now(), 'Iteration', iter, 'fwavacc', \
    (freq[freq > 0] * iu[freq > 0]).sum())