Research Project

All norms defination: https://rorasa.wordpress.com/2012/05/13/l0-norm-l1-norm-l2-norm-l-infinity-norm/

(1) Scipy function https://docs.scipy.org/doc/scipy-0.18.1/reference/generated/scipy.optimize.linear_sum_assignment.html (2) munkres. http://software.clapper.org/munkres/api/index.html

0.MOT challenge    https://motchallenge.net/ 1.FBMS(video segmentation benchmark) https://lmb.informatik.uni-freiburg.de/resources/datasets/ 2.http://videonet.team/

1.  without gt mask (1)in mot.py class mot ( imdb ): self .mask_size = cfg.TRAIN.MASK_SIZE #MRCNN $x=28 self .binary_thresh = 0.4 #MRCNN (2) def _load_mot_annotation ( self , index ): seg_mask_inds = np.zeros((num_objs, 2 ), dtype =np.uint32) #MRCNN # save image index and object index in this image ### MRCNN (need to be tested) if cfg.TRAIN.MASK_REG: #index has form: index = "MOT16-02_000003" --> has to parse into integer number index_t=index.strip() arr = index_t.split( '_' ) index_t2= int (arr[ 0 ][ 6 :]+arr[ 1 ]) ##tranfer MOT16-02_000003 --> 2000003 seg_mask_inds[ix, : ] = [index_t2, ix+ 1 ] #instance count from 1 ### MRCNN-seg im_size=tree.find( 'size' ) im_width= int (im_size.find( 'width' ).text) #print 'im_width:',im_width im_height= int (im_size.find( 'height' ).text) #print 'im_height:',im...

1.MOTA: the multiple object tracking accuracy      paper(1)附公式：(2) 2.MOTP: thei multiple object tracking precision 3.FAF: 3.GT No. of groundtruth trajectories. 4.MT MT% Mostly tracked: Percentage of GT trajectories which are covered by tracker output for more than 80% in length. 5.ML 5.ML% Mostly lost: Percentage of GT trajectories which are covered by tracker output for less than 20% in length. The smaller the better. 5. PT% Partially tracked: 1.0-MT-ML. 6.FP: false postive 7.FN: false negtive 8.IDsw  IDS ID switches: The total of No. of times that a tracked trajectory changes its matched GT identity. The smaller the better. 9.Frag Frag Fragments: The total of No. of times that a groundtruth trajectory is interrupted in tracking result. The smaller the better. 10.IoU 12. Recall (Frame-based) correctly matched objects / total groundtruth objects. 13： Precision (Frame-based) correctly matched objects / total output objects. FA/Frm...

1. reference link: https://github.com/rbgirshick/py-faster-rcnn 2. need copy the 'makefile.config'  file to the directory ''/home/uni/projects/maskrcnn/fasterrcnn/py-faster-rcnn/caffe-fast-rcnn" 3. the 4th step:     when using command:  make -j8 && make pycaffe     lhdf5 error      solution is:  change 'Makefile' file as follows: 删除--- LIBRARIES += glog gflags protobuf boost_system boost_filesystem m hdf5_hl hdf5 增加+++ LIBRARIES += glog gflags protobuf boost_system boost_filesystem m hdf5_serial_hl hdf5_serial 4. Run the demo " cd $FRCN_ROOT ./tools/demo.py " error (1) lack easydict solution: $  sudo pip install easydict error(2) lack cv2 solution: $ sudo apt-get install python-opencv 5. Usage:   cd $FRCN_ROOT ./experiments/scripts/faster_rcnn_alt_opt.sh [GPU_ID] [NET] [--set ...]    ./experiments/scripts/faster_rcnn_alt_opt.sh 0 VGG16 pascal_voc 6. error  and...

1.常用命令：    clear   % 清空变量窗口    clc      %　清空命令窗口 2. 创建array  x=[1 2 3]    % 1行3列的数组  x=[1;2;3]    % 3行1列的数组    x=[1 2 3] 和 x=1:3  等价  x=1:0.5:2     % 输出1 1.5 2  x=linspace(1,2,3)  % 输出 1 1.5 2 ，注意函数的参数之间是逗号隔开  x=x'       %对行向量进行转置  x=(5:2:9)'   % 输入结果为列向量  5  7  9  x=rand(4)   % 生成一个4*4的随机矩阵  x=rand(2,3)  %生成一个2*3的随机矩阵  x=zeros(2,3)  %生成一个2*3的零矩阵 3. 文件操作   save foo x    % 将变量x存入到文件foo.mat中   load foo.mat  % 从文件夹加载（读取）文件foo.mat到工作区 4.导入.txt，.png等文件     直接点击菜单栏的“导入数据” 5. 操作array(矩阵）    x=data(3,5)    %取出矩阵data的第五行、第三列元素并赋值给变量x    x=data(end, 5)  %取出矩阵data最后一行、第三列元素并赋值给变量x    x=data(end-1, 5)  %取出矩阵data倒数第二行、第三列元素并赋值给变量x    row2=data(2,:)   %取出矩阵dat...