Hackathon 2019

The hackathon will run from lunchtime 3rd July to end of the working day 4th July at Homerton College. While it won't be a competitive event, it will offer you the chance to:

• use (and improve!) your skills to work on interesting problems relating to computer vision, transport and the built environment
• contribute to a body of work that seeks to benefit population health
• meet, work with and learn from other people with similar interests
• get time and space away from your normal place of work to focus on something slightly different

We look forward to seeing you there!

Sign up to the hackathon here Registration now closed!

FAQs can be found here

Example Hackathon Challenges:

Problem Statement. Currently, usage levels of different transport modes is often estimated from travel survey data. It might be possible to make use of novel data resources such as Google Street View to estimate how people travel in particular locations. This would be particularly useful in lower and middle income countries where availability of survey data is limited.

Initial work has been done to suggest that data from novel resources could give useful estimates. However, this required manual assessment of thousands of images. Computer vision techniques could eliminate the manual work by detecting and counting objects such as cars and bikes.

Goal. To explore algorithmic approaches for detecting and counting objects relating to transport in data sets similar to those available via Google Street View

Reference paper: Goel R et al. (2018) Estimating city-level travel patterns using street imagery: A case study of using Google Street View in Britain.

Data set: TBC

Problem Statement: Object recognition is a challenging and active problem in computer vision. Whilst major progress has been reported using natural scenes, the development with remote sensing imagery has been considerably slower. This is mainly because of the significant variation in terms of shape, scale and orientation of the objects when viewed from above the Earth's surface. Remote sensing images could be used for estimating volume of larger vehicles on different kinds or roads, amount of space used for different transport modes, and assessing safety of road layouts.

Several works for object detection on aerial data exist, motivated by the state of the art results in deep learning for object recognition. These have been focussed on fine-tuning pre-trained models on well-known datasets such as MSCOCO. It has been pointed out in [Xia, 2018] that “the task of object detection in aerial images is distinguished from the conventional object detection”. This has motivated the development on large-scale datasets for aerial data such as DOTA [Xia, 2018].

Goal: To design a multi-class object detector.

Reference paper: Xia et al (2018), DOTA: A Large-scale Dataset for Object Detection in Aerial Images

Data set: DOTA data can be found here