Congratulations to Christoforos Charalambous (BICV Group), who won the prize for “Best Poster Presentation” at the 6th IET Conference on Imaging for Crime Detection and Prevention, held in London, 15-17 July 2015. This study – at the boundary between biomechanics and biometrics – investigates how viewing angle affects the performance of identity recognition using a person’s gait, particularly through the use of joint kinematics. Chris visualised how the accuracy of identity recognition varies with the spatial relationship between a security camera and the trajectory of a person captured using that camera. In addition, he showed how the accuracy of model-based gait recognition depends on the elevation angle of the camera, information which is likely to be useful for companies installing surveillance equipment. Chris’s prize was sponsored by the IEEE.
We’re just coming to the end of the 2-day NVIDIA CUDA Workshop, organised via the Deep Learning Network. Our instructor, Dr Anthony Morse (Plymouth University), covered a breadth of topics ranging from an introduction to CUDA, through optimisation to multi-GPU computing. He was able to draw from years of practical usage to cover some of the intricacies of CUDA that isn’t readily available through documentation and online courses.
After many hours of building Kai and Jose have finished constructing three new servers. The identical builds make it easier to transfer software between the machines, and also to investigate distributed computation for computer vision and machine learning. As scaling up becomes infeasible, the alternative is scaling out.
Each machine has a 8-core CPU, 64GB of RAM and 4 NVIDIA GPUs for General-Purpose computing on Graphics Processing Units (GPGPU). With such hardware we can make use of both parallelisation and virtualisation, the former enabling the training of large machine learning models and the latter allowing multiple users to trial different application environments on the same physical machine. The servers will see their first major test in the near future with the upcoming NVIDIA CUDA Workshops.
Our paper “Appearance-based indoor localization: A comparison of patch descriptor performance” is currently in Press. We will post the open-access final version soon, but in the mean time you can download the author accepted manuscript from arxiv.org: http://arxiv.org/pdf/1503.03514v1.pdf
As part of our commitment to both open source and reproducible research, BICV has started utilising Docker to create development environments for research into machine learning and computer vision. These environments consist of combinations of operating system + libraries + application environments, where OS/library dependencies are largely resolved. From a practical perspective this reduces the headache of compilation problems that many users struggle with daily. From a scientific perspective it allows code to be run in reproducible environments.
A Dockerfile can automate the building of Docker images. By making these publicly available on our Bitbucket repository, the creation of Docker images is not only transparent, but allows for further Docker images to be built on top of existing ones.
This code implements dense descriptors as described in the ECCV paper “Spatio-chromatic Opponent Features” by Alexiou and Bharath. The code is written in Matlab, but this release includes support for CUDA-acceleration with an appropriate GPU. A pre-print of the paper is available for download from the PDF link below:
and the code is available in the Zip package:
or from BitBucket.
We are pleased to announce that our paper “Associating locations from wearable cameras” by Jose Rivera-Rubio et al. has been accepted for BMVC 2014, Nottingham. BMVC is the fourth highest ranked conference in computer vision, with a rejection rate of 70%. The publication is closely related to the RSM dataset project, so you can read more here.
We are pleased to announce that our paper “Spatio-Chromatic Opponent Features” by Ioannis Alexiou and Anil Bharath has been accepted for ECCV 2014, Zurich. ECCV is the third highest ranked conference in computer vision, with a rejection rate of 75%. The code is now available on BitBucket, and a PDF preprint is available below: