Adjustable bounding boxes
FINISHED
- Task 1: Finished implementing adjustable bounding boxes feature
TO-DO
- Task 1: Support for Darknet
- Task 1: Add support to change class of wrongly detected objects
First part of week 6
Till Thursday time was spent to implement Adjustable bounding boxes feature and
neither of the approach suggested last week were used , i.e. neither
rectRegions.cpp nor Sample.cpp were used to Detect the mouse clicks, as
there were few issues while trying to accces the required parameters. So on
searching again for the optimum file to edit , Massinferencer.h was found.
First I tried updating a few parameters like the boundary box height, width and it’s
position to check if evrything was working good or not ,and thankfully it went
well. After this rest was straight forward, write the callback functions for
different types of MouseClick events and adjust the rectangles accordingly.
Results
Second part of week 6
To change the class names of wrongly classified objects is the task for next 1 week. The GUI for the same has been made and the update rule has also been written as of 7th July , but on closing the Qt window after selecting the required class , I’m encountering Segementation Fault error . And during today’s meeting I was asked to go through DatasetWriters code once so as to get better idea of the Suite is storing the inferenced results. And only during this week’s meeting I came across the main motive behind the creation of this tool and it was super cool, i.e. this tool is used to create Datasets for object detections in different required formats. Till now I thought the main use was to deploy the already trained models on different kinds of inputs, but that was only extra tasks this could perform , at the core this was made to generate training datasets for DeepLearning. Now currently work has been kept on hold for a day or two to go through different available types of DatasetWriters. Also am feeling super excited :)