mmrotate自定义数据集安装部署训练测试

环境

Ubuntu18.04
cuda10.2
python3.7
2080s

一、安装

conda create -n mmrotate python=3.7
conda activate mmrotate

#1、 安装pytorch 和 cuda
# 建议在官网根据cuda版本安装 https://pytorch.org/get-started/locally/
conda install pytorch torchvision torchaudio cudatoolkit=10.2 -c pytorch

#2、安装mmcv
# 安装完成后pip list看一下 torch的版本 方便安装对应版本的mmcv
# 我安装的是cuda10.2和torch1.11.0 
pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu102/torch1.11.0/index.html

#3、安装mmdetection
pip install mmdet

#4、从github上 clone mmrotate
git clone https://github.com/open-mmlab/mmrotate

#5、安装mmrotate并编译
cd mmrotate
pip install -r requirements.txt
pip install -V -e.

二、测试

# 1、下载pth文件
# 在demo目录的image_demo.py文件有该文件的使用说明
# 下载官方提供的pth文件
wget -P checkpoint https://download.openmmlab.com/mmrotate/v0.1.0/oriented_rcnn/oriented_rcnn_r50_fpn_1x_dota_le90

#2、在mmrotate目录下进行测试，执行如下命令
python demo/image_demo.py  demo/demo.jpg   configs/oriented_rcnn/oriented_rcnn_r50_fpn_1x_dota_le90.py checkpoint/oriented_rcnn_r50_fpn_1x_dota_le90-6d2b2ce0.pth

3、若弹出如下结果则说明环境安装正常

三、自定义数据集制作

mmrotate中所使用的数据集是dota格式的
DOTA标签格式

x1, y1, x2, y2, x3, y3, x4, y4, category: 目标名字 difficult：表示标签检测的难易程度 （1表示困难，0表示不困难）

x1,y1为左上角的坐标，然后顺时针排列4个坐标

1、使用rolabelimg工具标注数据
标注文件xml文件如下

2、生成DOTA数据集的标签格式

import os
import xml.etree.ElementTree as ET
import math
import cv2 as cv

def voc_to_dota(xml_path, xml_name):
    txt_name = xml_name[:-4] + '.txt'
    txt_path = xml_path + '/txt_label'
    if not os.path.exists(txt_path):
        os.makedirs(txt_path)
    txt_file = os.path.join(txt_path, txt_name)
    file_path = os.path.join(xml_path, file_list[i])
    tree = ET.parse(os.path.join(file_path))
    root = tree.getroot()
    # print(root[6][0].text)
    image_path = 'data/dota/images/'
    out_path = 'data/dota/outputImg/'
    filename = image_path + xml_name[:-4] + '.jpg'
    img = cv.imread(filename)
    with open(txt_file, "w+", encoding='UTF-8') as out_file:
        # out_file.write('imagesource:null' + '\n' + 'gsd:null' + '\n')
        for obj in root.findall('object'):
            name = obj.find('name').text
            difficult = obj.find('difficult').text
            # print(name, difficult)
            robndbox = obj.find('robndbox')
            cx = float(robndbox.find('cx').text)
            cy = float(robndbox.find('cy').text)
            w = float(robndbox.find('w').text)
            h = float(robndbox.find('h').text)
            angle = float(robndbox.find('angle').text)
            # print(cx, cy, w, h, angle)
            p0x, p0y = rotatePoint(cx, cy, cx - w / 2, cy - h / 2, -angle)
            p1x, p1y = rotatePoint(cx, cy, cx + w / 2, cy - h / 2, -angle)
            p2x, p2y = rotatePoint(cx, cy, cx + w / 2, cy + h / 2, -angle)
            p3x, p3y = rotatePoint(cx, cy, cx - w / 2, cy + h / 2, -angle)

            # 找最左上角的点
            dict = {p0y:p0x, p1y:p1x, p2y:p2x, p3y:p3x }
            list = find_topLeftPopint(dict)
            #print((list))
            if list[0] == p0x:
                list_xy = [p0x, p0y, p1x, p1y, p2x, p2y, p3x, p3y]
            elif list[0] == p1x:
                list_xy = [p1x, p1y, p2x, p2y, p3x, p3y, p0x, p0y]
            elif list[0] == p2x:
                list_xy = [p2x, p2y, p3x, p3y, p0x, p0y, p1x, p1y]
            else:
                list_xy = [p3x, p3y, p0x, p0y, p1x, p1y, p2x, p2y]

            # 在原图上画矩形 看是否转换正确
            cv.line(img, (int(list_xy[0]), int(list_xy[1])), (int(list_xy[2]), int(list_xy[3])), color=(255, 0, 0), thickness= 3)
            cv.line(img, (int(list_xy[2]), int(list_xy[3])), (int(list_xy[4]), int(list_xy[5])), color=(0, 255, 0), thickness= 3)
            cv.line(img, (int(list_xy[4]), int(list_xy[5])), (int(list_xy[6]), int(list_xy[7])), color=(0, 0, 255), thickness= 2)
            cv.line(img, (int(list_xy[6]), int(list_xy[7])), (int(list_xy[0]), int(list_xy[1])), color=(255, 255, 0), thickness= 2)

            data = str(list_xy[0]) + " " + str(list_xy[1]) + " " + str(list_xy[2]) + " " + str(list_xy[3]) + " " + \
                   str(list_xy[4]) + " " + str(list_xy[5]) + " " + str(list_xy[6]) + " " + str(list_xy[7]) + " "
            data = data + name + " " + difficult + "\n"
            out_file.write(data)
       # cv.imwrite(out_path + xml_name[:-4] + '.jpg', img)


def find_topLeftPopint(dict):
    dict_keys = sorted(dict.keys())  # y值
    temp = [dict[dict_keys[0]], dict[dict_keys[1]]]
    minx = min(temp)
    if minx == temp[0]:
        miny = dict_keys[0]
    else:
        miny = dict_keys[1]
    return [minx, miny]


# 转换成四点坐标
def rotatePoint(xc, yc, xp, yp, theta):
    xoff = xp - xc
    yoff = yp - yc
    cosTheta = math.cos(theta)
    sinTheta = math.sin(theta)
    pResx = cosTheta * xoff + sinTheta * yoff
    pResy = - sinTheta * xoff + cosTheta * yoff
    # pRes = (xc + pResx, yc + pResy)
    # 保留一位小数点
    return float(format(xc + pResx, '.1f')), float(format(yc + pResy, '.1f'))
    # return xc + pResx, yc + pResy


if __name__ == '__main__':
    root_path = 'data/dota/xml'
    file_list = os.listdir(root_path)
    for i in range(0, len(file_list)):
        if ('.xml' in file_list[i]) or ('.XML' in file_list[i]):
            voc_to_dota(root_path, file_list[i])
            print('----------------------------------------{}{}----------------------------------------'
                  .format(file_list[i], ' has Done!'))
        else:
            print(file_list[i] + ' is not xml file')

3、划分训练集验证集测试集
划分后的目录文件结构

python mmrotate-main/tools/data/dota/split/img_split.py --base_json mmrotate-main/tools/data/dota/split/split_configs/ss_train.json

4、切分图像
使用mmrotate中自带的tools工具，将数据进行裁减，并重新设为png格式的图像。
修改/home/cv/mmrotate/tools/data/dota/split/split_configs中的json文件，修改每个文件中的，如下几个路径即可。

分别使用脚本进行裁减

 1411  python img_split.py --base-json split_configs/ss_train.json 
 1412  python img_split.py --base-json split_configs/ss_test.json 
 1413  python img_split.py --base-json split_configs/ss_trainval.json 
 1414  python img_split.py --base-json split_configs/ss_val.json 

四、修改训练参数

1、修改mmrotate/mmrotate/datasets的dota.py文件中的类别

2、修改mmrotate/configs/rotated_faster_rcnn/rotated_faster_rcnn_r50_fpn_1x_dota_le90.py中的类别数量

修改rotated_faster_rcnn_r50_fpn_1x_dota_le90.py中的类别数量

修改mmrotate/configs/base/datasets中的数据集路径

修改中mmrotate/configs/base/schedules/schedule_1x.py中的参数

修改mmrotate/configs/base/default_runtime.py中的参数

3、修改train中的参数
主要修改–config和–word-dir

python tools/train.py 

提示如下信息为在训练

如果遇到 Cuda out of memory错误：可将 mmrotate-main/configs/base/datasets/dotav1.py文件中的samples_per_gpu 和 workers_per_gpu 改小一点。

六、测试

修改tools/tests.py

python tools/test.py --show

参考资料
https://blog.csdn.net/qq_43581224/article/details/123838415?ops_request_misc=%257B%2522request%255Fid%2522%253A%2522165447944816782390592349%2522%252C%2522scm%2522%253A%252220140713.130102334…%2522%257D&request_id=165447944816782390592349&biz_id=0&utm_medium=distribute.pc_search_result.none-task-blog-2_alltop_positive~default-1-123838415-null-null.142^v11pc_search_result_control_group,157^v13new_style1&utm_term=mmrotate&spm=1018.2226.3001.4187