Explanation in Portuguese

Tools to annotate images

• ChangeImagesMaxSize
• ExtractImagesFromVideo
• CreateBoundingBoxes
• ExportToClassification
• ExportToCSV
• ExportToPascal (🔼 2019-03-14)
• Generate TFRecord - API Tensorflow (🔼 2019-03-19)
• Import from txt ground truth - format Wider Face (🔼 2019-03-14)
• Convert CityscapeMask annotatiton to XML Pascal.py (🔼 2019-06-10)

Tools to measures

• mAP

Citation

@misc{marcelo_cabral_ghilardi_2019_2604909,
  author       = {Marcelo Cabral Ghilardi},
  title        = {kabrau/PyImageRoi: Tools to annotate images},
  month        = mar,
  year         = 2019,
  doi          = {10.5281/zenodo.2604909},
  url          = {https://doi.org/10.5281/zenodo.2604909}
}

ChangeImagesMaxSize

A tool to change images max size into folder

RUN

usage: ChangeImagesMaxSize.py [-h] -p PATH -s MAXSIZE

optional arguments:
  -h, --help            show this help message and exit
  -p PATH, --path PATH  images path
  -s MAXSIZE, --maxSize MAXSIZE
                        resize image (width ou height) to MAX SIZE

ExtractImagesFromVideo

A tool to extract images from videos

The images name is same video name with a sequential number, e.g.:

• video: VID-20170817-WA0003.mp4
• images: VID-20170817-WA0003-(F00001).txt, VID-20170817-WA0003-(F00002).txt, VID-20170817-WA0003-(F00003).txt

RUN

usage: ExtractImagesFromVideo.py [-h] -p VIDEOSPATH -o OUTPUTPATH [-f FPS]
                                 [-n]
                                 {portrait,landscape}

positional arguments:
  {portrait,landscape}  portrait (default) or landscape

optional arguments:
  -h, --help            show this help message and exit
  -p VIDEOSPATH, --videosPath VIDEOSPATH
                        videos input path
  -o OUTPUTPATH, --outputPath OUTPUTPATH
                        images output path
  -f FPS, --fps FPS     extract frames por seconds
  -n, --new             Extracts only from videos without extraction (new
                        video)              Extracts only from videos without extraction (new videos) <br>

CreateBoundingBoxes

A tool to Labeling object bounding boxes or ROI (Region of interest) in images
(adjusts the displayed image size to the screen size)

• multiple boxes per image
• multiple classes per image

The regions are saved in a text file with same name of image file, e.g.

IMG-20170807-WA0001.jpg
IMG-20170807-WA0001.txt
locaisvstdss.jpg
locaisvstdss.txt
phpiqa6ae.752.502.s.jpg
phpiqa6ae.752.502.s.txt

** Important: ** Do not put two pictures with the same name and different extensions in the same folder.

Each line of text file is a one region

class_number x1 y1 width height image_width image_height

Example:

1 426 679 55 99 1080 1920
1 440 839 30 59 1080 1920

RUN

usage: CreateBoundingBoxes.py [-h] -p PATH [-f] [-c CLASS]
                              [-className [CLASSNAME [CLASSNAME ...]]]

optional arguments:
  -h, --help            show this help message and exit
  -p PATH, --path PATH  images path
  -f, --first           starts on the first image (default: Jump to first
                        image without label)
  -c CLASS, --class CLASS
                        class number started (default = 0)
  -className [CLASSNAME [CLASSNAME ...]]
                        class name list (0..9 positions, max 10), e.g.
                        -className dog cat

USAGE

Left Click mouse to start marking an area
Right Click mouse to remove last area
'0..9' change class to new boxe (aaccent key = 0, too)
'N' or 'space-bar' to next image
'P' to previus image
'Q' Exit

Example

python CreateBoundingBoxes.py -p ...\image -className cat plant

0 227 111 662 359 1024 576 
1 647 255 114 173 1024 576 
1 756 257 115 172 1024 576 
1 4 180 164 316 1024 576 

ExportToClassification

A tool to extract box from images and save imagebox to classification. Save the separated images by classes, each class in a subfolder.

RUN

usage: ExportToClassification.py [-h] -p PATH -d DEST

optional arguments:
  -h, --help            show this help message and exit
  -p PATH, --path PATH  images path
  -d DEST, --dest DEST  destination images path

Example

python ExportToClassification.py -p ...\image -f ...\image

..\.\tmp\0 MyCat-0.jpg [227, 111, 889, 470, '0'] (359, 662, 3)
..\.\tmp\1 MyCat-1.jpg [647, 255, 761, 428, '1'] (173, 114, 3)
..\.\tmp\1 MyCat-2.jpg [756, 257, 871, 429, '1'] (172, 115, 3)
..\.\tmp\1 MyCat-3.jpg [4, 180, 168, 496, '1'] (316, 164, 3)

ExportToCSV

A tool to create a cvs file with de bounding boxes

note: Use 1 classname only

RUN

usage: ExportToCSV.py [-h] -p PATH -c CVS_FILE

optional arguments:
  -h, --help            show this help message and exit
  -p PATH, --path PATH  images path
  -c CVS_FILE, --cvs_file CVS_FILE
                        cvs file

Example

python ExportToCSV.py -p E:\Datasets\pedestrian_signal\images\test -c E:\Datasets\pedestrian_signal\images\test.csv

filename,width,height,class,xmin,ymin,xmax,ymax
16431531.jpg,640,426,sinaleira,317,92,345,140
16431540.jpg,640,426,sinaleira,449,106,475,148
17074299.jpg,620,412,sinaleira,566,199,586,228
19156210.jpg,620,412,sinaleira,181,112,206,161
20170701_105311.jpg,768,1024,sinaleira,323,424,352,471
20170701_105442.jpg,768,1024,sinaleira,311,412,346,463

ExportToPascal

TOOL to Create a XML files (PASCAL FORMAT)

RUN

usage: ExportToPascal.py [-h] -p PATH -a ANNPATH 
                              [-className [CLASSNAME [CLASSNAME ...]]]

optional arguments:  
  -h, --help            show this help message and exit  
  -p PATH, --path PATH  images path  
  -a ANNPATH, --annpath ANNPATH  
                        annotation path  
  -className [CLASSNAME [CLASSNAME ...]]
                        class name list (0..9 positions, max 10), e.g.
                        -className dog cat

Example

python ExportToPascal.py -p "E:\Datasets\pedestrian_signal\images" -a "E:\Datasets\pedestrian_signal\images.ann_gostop" -className go stop off

Ps: At the end, it shows total images and classes

ExportToPascal.5971774

Specific converter for PedestrianLights dataset available at: http://www.uni-muenster.de/PRIA/en/forschung/index.shtml
TOOL to Create a XML files (PASCAL FORMAT)

RUN

usage: usage: ExportToPascal.5971774.py [-h] -p PATH -o GTFILE -a ANNPATH   

optional arguments:  
  -h, --help            show this help message and exit  
  -p PATH, --path PATH  images path  
  -o GTFILE, --gtfile GTFILE  
                        original ground truth file  
  -a ANNPATH, --annpath ANNPATH  
                        annotation path  

Example

python ExportToPascal.5971774.py -p "E:\Datasets\pedestrianlights-5971774\pedestrianlights\download\imagesequences\01" -o "E:\Datasets\pedestrianlights-5971774\pedestrianlights\download\imagesequences\01\groundtruth.txt" -a "E:\Datasets\pedestrianlights-5971774\pedestrianlights\download\imagesequences\01.ann.GoStop"

Ps: At the end, it shows total images and classes

Generate TFRecord - API Tensorflow

• First, convert from PASCAL to CSV, use: ExportPascal2csv.py
• Second, generate TFRecord, use: Generate_TFRecord.py

RUN

usage: ExportPascal2csv.py [-h] -p PATH -o OUTPUT [-a]

optional arguments:
  -h, --help            show this help message and exit
  -p PATH, --path PATH  annotations path
  -o OUTPUT, --output OUTPUT
                        csv output file
  -a, --addPathCol      add path collumn  

usage: Generate_TFRecord.py [-h] -c CSV_INPUT -o OUTPUT_PATH [-i IMAGES_PATH]

optional arguments:
  -h, --help            show this help message and exit
  -c CSV_INPUT, --csv_input CSV_INPUT
                        Path to the CSV input
  -o OUTPUT_PATH, --output_path OUTPUT_PATH
                        Path to output TFRecord
  -i IMAGES_PATH, --images_path IMAGES_PATH
                        Path for Images, If dont have into CSV

Example

python ExportPascal2csv.py -p E:\datasets\FaceDataset\Wider\WIDER_train\train.ann -o E:\datasets\FaceDataset\Wider\WIDER_train\train.csv -a

python Generate_TFRecord.py --csv_input=E:\datasets\FaceDataset\Wider\WIDER_train\train.csv --output_path=E:\datasets\FaceDataset\Wider\WIDER_train\train.record

mAP

Tool to calc mAP in Object Detection

Returns in the console the mAP values
And inside each method folder create a subfolder named _chart with a pdf of the Precision Recall method

You need a folder with Pascal VOC annotations, parameter --annpath

And a root folder with one ou more methods result folder, parameter --resultpath e.g:

c:\results <= root
c:\results\faster <= method result folder
c:\results\ssd <= method result folder
c:\results\yolo <= method result folder

Inside each method result folder, you need results files by classes
e.g:
cat.txt
dog.txt
mouse.txt

In the results files by class, the content is in this format:
filename confidence x1 y1 x2 y2
e.g:
file001 0.99862 441.5266 429.1418 504.2249 548.0778
file005 0.99757 466.8359 433.5500 531.3656 545.9105
file007 0.95728 495.3467 440.6576 554.5069 558.3262

RUN

usage: mAP.py [-h] -a ANNPATH -r RESULTPATH -c [CLASS [CLASS ...]] -i IOU [-v [VERBOSE]]  
  
  
optional arguments:  
  -h, --help            show this help message and exit   
  -a ANNPATH, --annpath ANNPATH  
                        Pascal VOC annotation path  
  -r RESULTPATH, --resultpath RESULTPATH  
                        Path of method results  
  -c [CLASS [CLASS ...]], --class [CLASS [CLASS ...]]  
                        list of class, e.g. --classes dog cat mouse  
  -i IOU, --IOU IOU     IOU confidence threshold, e.g. 0.5  
  -v [VERBOSE], --verbose [VERBOSE]  
                        show verbose  

Example

python mAP.py -a E:\Datasets\signal\test.ann.GoStop -r E:\GitHub\PedestrialTrafficLight\accurace_calc\results\3C\ -c cat dog mouse -i 0.5

Import from txt ground truth - format Wider Face

Wider Face - A Face Detection dataset to Benchmark http://mmlab.ie.cuhk.edu.hk/projects/WIDERFace/index.html

The format of txt ground truth.

File name
Number of bounding box
x1, y1, w, h, blur, expression, illumination, invalid, occlusion, pose

RUN

usage: ImportFromTxtGT_01.py [-h] -p PATH -f FILE

optional arguments:
  -h, --help            show this help message and exit
  -p PATH, --path PATH  images path
  -f FILE, --file FILE  ground truth file

Example

python importFromTxtGT_01.py -p E:/datasets/FaceDataset/Wider/WIDER_train/images/ -f E:/datasets/FaceDataset/Wider/wider_face_split/wider_face_train_bbx_gt.txt

Import from txt ground truth - format Wider Face

Convert CityscapeMask annotatiton to XML Pascal.py

Open code CityscapeMask2Pascal.py and set folders and classes

RUN

python CityscapeMask2Pascal.py