cunyizhang的博客

# 目标检测中，将pascal格式的数据标签Annotations按照train.txt、val.txt分成对应的文件夹# 用在retinanet中的训练集和验证集 import osimport shutildef split_file(src_file_path, dst_file, dst_file_path):    file = open(dst_file, 'r...

前言在用目标检测做分类的时候,模型不会判断物体属性,只是根据学习来判断可能是什么标签,并给出bbox. 但是研究内容主要是针对人,所有桌子等其他干扰必须去除.想到是否可以对所有样本数据新增一个label="person".意思就是对原来所有的xml文件中,有object的label翻倍并修改新增的object的label为"person".代码import osimport num...

import cv2import numpy as np import xml.dom.minidomimport osimport argparseimg_path = 'images/'anno_path = 'annotations/'cut_path = 'cut/'imagelist = os.listdir(img_path)for image in imag...

转自:目标检测系列一：如何制作数据集？txt to coco# -*- coding: UTF-8 -*-import cv2import jsonimport sys# process bardef process_bar(count, total, status=''): bar_len = 60 filled_len = int(round(bar_...

转自:opencv+python3.6图像旋转、并保持图像像素大小不变、不被裁剪（旋转90/180/270度）import cv2import numpy as npimport os, shutildef rotate_bound(image, angle): # grab the dimensions of the image and then determine th...

#opencv批量泊松融合import cv2import numpy as npimport ossrc_path = "cut_1/"save_path = "mixup_1/"dst = cv2.imread("beijing1.jpg")a = dst.shapeH=a[0]W=a[1]print("H",H)print("W",W)imagelist = os...

python 文件open函数 r、w、a模式

# -*- coding: UTF-8 -*-import cv2import jsonimport sys# process bardef process_bar(count, total, status=''): bar_len = 60 filled_len = int(round(bar_len * count / float(total))) p...