Projects

Research

Light-Weight RefineNet for Real-Time Semantic Segmentation
V.Nekrasov, C.Shen, I.Reid
British Machine Vision Conference, 2018
code / paper
Re-thinking an encoder-decoder based segmentation network into the one able to attain high performance with the real-time inference
Real-Time Joint Semantic Segmentation and Depth Estimation Using Asymmetric Annotations
V.Nekrasov, T.Dharmasiri, A.Spek, T.Drummond, C.Shen, I.Reid
paper
A simple approach of performing joint segmentation, depth and surface normals estimation in real-time with SOTA results using a single model
Diagnostics in Semantic Segmentation
V.Nekrasov, C.Shen, I.Reid
paper / supp
Evaluating SOTA segmentation networks in terms of their sensitivity to object properties and uncertainty measures
Global Deconvolutional Networks for Semantic Segmentation
V.Nekrasov, J.Ju, J.Choi
British Machine Vision Conference, 2016
code / paper
Learnable matrix-based method for global interpolation inside segmentation networks

Re-implementations

DeepLab-v1
TensorFlow
code
Re-implementation of DeepLab-v1 (LargeFOV) in TensorFlow
DeepLab-v2
TensorFlow
code
Re-implementation of DeepLab-v2 (ResNet-101) in TensorFlow
DeepLab-v3+
PyTorch
code
Conversion of DeepLab-v3+ pre-trained weights from TensorFlow into PyTorch

Motley

Bio-Data Programming
code / shiny1 / shiny2
Shiny-based applications developed for the course on biomedical data programming - based on the NCI-60 cancel cell line panel dataset
EPL-16 Away Fixtures Map
code / shiny
Mapping of away fixtures in English Premier League Season ‘16-‘17 (first half) using leaflet and d3
GPU-server workload visualisation
code
Simple visualisation of workload on a GPU server in d3.js