mAP&AP算法 https://www.zhihu.com/question/53405779
1.代碼來自於https://github.com/matterport/Mask_RCNN
訓練自己的數據參考博客https://blog.csdn.net/qq_29462849/article/details/81037343#commentBox
批量重命名文件,注意,如果文件本身就是json格式,需要先轉化爲png,再變爲json,因爲這是處於一個for 循環中,所以之前的又被重複使用了
i=1; for x in *; do mv $x $i.json; let i=i+1; done
這裏說明一下用labelme製作的數據集出現的問題
因爲labelme升級以後,用命令行labelme_json_to_dataset+空格+文件名稱.json就產生不了能用的圖片
因此,對於labelme_json裏面的文件,應該用如下py文件生成
import argparse
import json
import os
import os.path as osp
import warnings
import PIL.Image
import yaml
from labelme import utils
def main():
warnings.warn("This script is aimed to demonstrate how to convert the\n"
"JSON file to a single image dataset, and not to handle\n"
"multiple JSON files to generate a real-use dataset.")
parser = argparse.ArgumentParser()
parser.add_argument('json_file')
parser.add_argument('-o', '--out', default=None)
args = parser.parse_args()
json_file = args.json_file
print("json_flie= ",json_file)
if args.out is None:
out_dir = osp.basename(json_file).replace('.', '_')
out_dir = osp.join(osp.dirname(json_file), out_dir)
else:
out_dir = args.out
if not osp.exists(out_dir):
os.mkdir(out_dir)
data = json.load(open(json_file))
print("json---load")
img = utils.img_b64_to_arr(data['imageData'])
label_name_to_value = {'_background_': 0}
for shape in data['shapes']:
label_name = shape['label']
if label_name in label_name_to_value:
label_value = label_name_to_value[label_name]
else:
label_value = len(label_name_to_value)
label_name_to_value[label_name] = label_value
# label_values must be dense
label_values, label_names = [], []
for ln, lv in sorted(label_name_to_value.items(), key=lambda x: x[1]):
label_values.append(lv)
label_names.append(ln)
assert label_values == list(range(len(label_values)))
lbl = utils.shapes_to_label(img.shape, data['shapes'], label_name_to_value)
captions = ['{}: {}'.format(lv, ln)
for ln, lv in label_name_to_value.items()]
lbl_viz = utils.draw_label(lbl, img, captions)
PIL.Image.fromarray(img).save(osp.join(out_dir, 'img.png'))
PIL.Image.fromarray(lbl).save(osp.join(out_dir, 'label.png'))
PIL.Image.fromarray(lbl_viz).save(osp.join(out_dir, 'label_viz.png'))
with open(osp.join(out_dir, 'label_names.txt'), 'w') as f:
for lbl_name in label_names:
f.write(lbl_name + '\n')
warnings.warn('info.yaml is being replaced by label_names.txt')
info = dict(label_names=label_names)
with open(osp.join(out_dir, 'info.yaml'), 'w') as f:
yaml.safe_dump(info, f, default_flow_style=False)
print('Saved to: %s' % out_dir)
if __name__ == '__main__':
main()
1.1以上代碼無需改動,放置在json文件夾中即可
1.2但是爲了批量處理,需要用到腳本文件
#!/bin/bash
# -*- coding: utf-8 -*-
for((i=1;i<280;i++))
do
python /home/tj816/mask-rcnn/train_data/json/json_to_dataset.py /home/tj816/mask-rcnn/train_data/json/${i}.json
done
python 後面第一個是py文件的位置,第二個是json圖片的位置,如1.json圖片的位置
1.3然後在自己的source activate tf的環境中,運行sh labelme2dataset.sh這個腳本文件,然後我們將得到需要的labelme_json文件
1.4用cpp將16位的png圖片改爲8位的png圖片,buff1是原圖片位置,buff2是目標圖片位置
#include <iostream>
#include <opencv2/opencv.hpp>
#include <string>
#include <stdio.h>
using namespace std;
using namespace cv;
int main(void){
char buff1[100]; char buff2[100];
for(int i=1;i<280;i++){
sprintf(buff1,"/home/tj816/mask-rcnn/train_data/labelme_json/%d_json/label.png",i);
sprintf(buff2,"/home/tj816/mask-rcnn/train_data/cv2_mask/%d.png",i);
Mat src;
//Mat dst;
src=imread(buff1,CV_LOAD_IMAGE_UNCHANGED);
Mat ff=Mat::zeros(src.rows,src.cols,CV_8UC1);
for(int k=0;k<src.rows;k++){
for(int kk=0;kk<src.cols;kk++){
int n=src.at<ushort>(k,kk);
ff.at<uchar>(k,kk)=n; }
}
imwrite(buff2,ff);
}
return 0;
}
1.5利用shell命令行進行編譯
附註:安裝opencv2 ,解決版本問題 https://blog.csdn.net/m0_37393514/article/details/87905645
解決CMake Error: The following variables are used in this project, but they are set to NOTFOUND.
Please set them or make sure they are set and tested correctly in the CMake files:
https://blog.csdn.net/u014613745/article/details/78310916
注意,第4、5步不用管。
g++ example.cpp `pkg-config opencv --libs --cflags opencv` -o example
1.6 ./example執行即可!
2.瞭解各個包中內容的含義:
2.1 demo.ipynb,利用預訓練模型權重+mask rcnn框架,隨機抽取一張圖片,進行檢測!
參數及含義:
BACKBONE:選取的網絡類型,如res101
BACKBONE_STRIDES,特徵圖相對於原圖縮小的倍數,如在resnet101中,C2的大小是原圖的1/4,C3是原圖的1/8,C4是的原圖1/16,C5是原圖的1/32,其和FPN輸出的P2-P5是對應的,P6在P5的基礎上進行了一個下采樣,因此P6是原圖的1/64(這樣可以決定每一層feature map最後的輸出尺寸)
BATCH_SIZE,每一個batch,包含多少張圖片
3.保存tensorboard的損失曲線爲圖片
若未安裝tensorboard,則需要先安裝tensorboard
conda install tensorboard
然後執行(shapes20190307T1102裝訓練出來權重的文件夾)
tensorboard --logdir=/home/tj816/Mask_RCNN-master/logs/shapes20190307T1102
將數據以csv的形式下載下來,再用ubuntu的excel來製作圖表即可!