Multi-Object Tracking with Transformers and DeepStream
DISCLAIMER: IF YOU SELECT TO DO THIS PROJECT YOU WILL BE GIVEN ACCESS TO AWS SPOT INSTANCE with GPU (ONLY IF YOU DON’T HAVE ONE).
Multi-Object Tracking (MOT) is a unique ability of humans and currently, under the headline of ’Probabilistic Reasoning over Time”, is being actively pursued by computer scientists that submit their findings to the MOT Challenges competition.
In this project you will be using this MOT17 challenge dataset to implement a State of The Art (SOTA) method called Trackformers as described in this paper and implemented in this repo.
Your team needs to select one of the two tasks bearing in mind that 10 bonus points will be given to teams implementing the segmentation tracking task.
Environment setup (10 points)
You will be given an AWS Cloud Development Kit script that will allow you to deploy and teardown an EC2 p2.2xlarge instance in AWS. You will need to work with the NVIDIA Deepstream docker and to download it you need an NVIDIA developer account.