I am about to make a vehicle detection model which is able to detect all types of vehicles like bikes, motorcycles, cars, buses and trucks and return the detection boxes. Unfortunately I did not find good dataset in some categories online, so I decided to make one for myself. My questions:
1- what sizes of images should I include? is it even important? (the images probably are going to be resized to 224*224*3 for model input)
2- how can I gather images that has good coverage over the vehicle type?
3- how can I use saliency maps to improve my understanding of what my dataset lacks in features?
Article On-Road Vehicle Detection: A Review
Article Methods for comparing scanpaths and saliency maps: Strengths...
1200x600 should be good enough. It really depends on how far/small you want to detect. It will all depend on the features you select.
I think this paper is very important
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5075645/
You may check KITTI video dataset. This is very popular dataset in autonomous driving and its challenges. Masoud Masoumi Moghadam
- Badges
- Science topic