As a part of team AGV’s task to slow down self driving cars upon encountering bumpers, I had been roaming around the campus clicking images of bumpers using my smartphone and used crude simple image editing to create binary images with the bumpers coloured white and the rest part of the images coloured in black. A lot of them were taken while cycling to class, or while coming back from the classes. The processing done using Krita was quite basic and crude, and the dataset may not be very well made, but I think it still is learn-able.
The original images were of ultraHD resolution and hence image processing and learning algorithms would have taken a lot of time to work with them. I batch-downscaled the images to 720p and a total of 110 images + 110 binary counterparts, each having 720p resolution are a part of the dataset.
I also have written a code that spews out negative and positive samples from the images, and puts them into separate folders. Most images turn out to be negatives, so some amount of manual deletion is required to make the generated image set learn-able.
I hope to make the convnet quickly and turn this dataset into an awesome computer vision application as soon as possible.