classification

Classification Label Generator Toolset

CLG-Toolset creates a classification label validation list for a given image database and labels set. The images in the dataset are assumed to have the correct labels in their metadata. The tools help to rename the dataset, resize the images, pad the images with a value(0-255) if they are resized to a square image to keep the aspect ratio, extract the labels from the metadata, and generate logs to indicate errors and mismatches in the dataset.

Prerequisites for running the Toolset

Linux

• python
• pil
• exiftool

sudo apt install libimage-exiftool-perl
sudo apt-get install python-pip
sudo apt-get install python-dev libjpeg-dev libfreetype6-dev zlib1g-dev
sudo pip install pil

Windows

• python
• pil
• exiftool
• qawk

Running the Toolset

Optional - rename the dataset to cleanup invalid filenames

Fix all file names in the input image folder by running the following command inside the image folder

ls | cat -n | while read n f; do mv "$f" "image-$n.jpg"; done

Image DataBase Creator

Run imageDataBaseCreator.py to create the image database with the required width, height, padding, and image name

python imageDataBaseCreator.py 	-d [input image directory] 	--- required 
 								-o [output image directory] --- required
 								-f [new image file name] 	--- required
 								-w [resize width] 			--- optional
 								-h [resize height] 			--- optional
 								-p [padding value] 			--- optional
 								-c [image start count] 		--- optional

Outputs

1. Output Image Directory - this folder contains all the images resized and renamed

2. fileName-val.txt - this is the classification label validation text file (fileName -- -f option )

imagename_1.JPEG 122
imagename_2.JPEG 928

3. fileName-scriptOutput - this folder contains all the logs and error files

• fileName-fileNameTanslation.csv
• fileName-fileNameWithErrors.csv
• fileName-fileNameWithLabels.csv
• fileName-invalidLabelsFile.csv
• fileName-multipleLabelsFile.csv

Scripts

This scripts folder has the following python scripts.

step-1.py

Run step-1.py to resize and rename your image to the required width and height, also allows the padding to keep the image resolution

python step-1.py 	-d [input image directory] 	--- required 
 					-o [output image directory] --- required
 					-f [new image file name] 	--- required 
 					-w [resize width] 		    --- optional
 					-h [resize height] 			--- optional
 					-p [padding value] 			--- optional

this script will resize and rename all your images and put them in the output folder you created.

step-2.py

Run step-2.py to extract all the tags and output a text file with the image name and all the tags associated with the image

python step-2.py 	-d [input image directory] 	--- required 
 					-f [tag_file_name.txt] 		--- required 

this script will output a CSV format image name & tags. The output file will be CSV_tag_file_name.txt

output example:

imagename.JPEG, tench, Tinca tinca (fileName.JPEG, tags)

step-3.py

Run step-3.py to create a usable image validation .txt with the image name and class number

python step-3.py 	-l [label.txt with 1000 labels without synset numbers] 	--- required (script-labels.txt from this project)
 					-t [CSV_tag_file_name.txt] 								--- required (output from step 2)

this script will generate an annie inference app usable data on the cmd/terminal use >> to val.txt for output

output example:

imagename.JPEG 0 (fileName.JPEG Label)

imageDataBaseCreator.py

Run imageDataBaseCreator.py to create the image database with the required width, height, padding, and image name

python imageDataBaseCreator.py 	-d [input image directory] 	--- required 
 								-o [output image directory] --- required
 								-f [new image file name] 	--- required 
 								-w [resize width] 			--- optional
 								-h [resize height] 			--- optional
 								-p [padding value] 			--- optional
 								-c [image start count] 		--- optional