iLABEL

The aim of this work is to semi-automatise labelling image process. We want to achieve this through:

speeding up the process
decrease the human error
make the process reproducible

Case study

Here we want to label a set of 1k images with some quality measures. Those labels are:

overall quality
light
growth
focus

Starting from this...

im_id	overall	light	...	focus
im1	NA	NA	...	NA
im2	NA	NA	...	NA
im3	NA	NA	...	NA

Practically, the user will be asked to:

load a set of images (specify the path)
visualise one image at time
input the value and submit
go to next image and repeat
save and eventually continue from this step

And we get back...

im_id	overall	light	...	focus
im1	0.1	0.1	...	0.3
im2	0.9	0.6	...	1
im3	0.7	0.6	...	0.5

In the backhand

When the user defines the list of images a database have to be created. When the user submit his values, they should be added to the images on the db.

An idea for the interface

Some test images to play with can be found here

To keep in mind (not mandatory at the beginning)

Number and name of the metrics could vary depending on the need

apollo994/iLABEL