Koustav Mullick
Computer Vision Researcher at Bosch Research
Bengaluru, Karnataka, India
Summary
👤
Senior🎓
Top SchoolKoustav Mullick is a Computer Vision researcher with 10 years of experience, currently driving applied research and activity management at Bosch Research in Bengaluru. He specializes in machine learning for 3D and volumetric vision, with hands-on expertise implementing and optimizing CUDA and PyTorch backend modules for volumetric convolution, pooling and unpooling. His contributions to foundational deep-learning libraries include performance-focused additions like VolumetricConvolutionMM and CUDA padding support, reflecting a strong systems-to-algorithm understanding. Trained at IIIT Hyderabad, he blends rigorous academic grounding with practical engineering that scales from research prototypes to production-grade kernels. Colleagues rely on him for bridging low-level GPU optimization and high-level generative/vision models, often uncovering performance wins that are not obvious from model design alone. Based in India, he pairs deep technical execution with program-level coordination across research initiatives.
10 years of coding experience
Master’s Degree, Master’s Degree at IIIT HyderabadHigh School, High School at St. Xavier's InstitutionBachelor’s Degree, Bachelor’s Degree at Heritage Institute of Technology, Kolkata
Github Skills (17)
pytorch10
convolutional-neural-networks10
machine-learning10
c1110
c1710
mask-rcnn10
deep-learning10
neural-network10
cuda10
faster-rcnn10
c-language9
lib9
matrix-multiplication9
cprogramming-language9
algorithms8
Programming languages (4)
Github contributions (5)
torch/cunn
Sep 2015 - Aug 2016
Role in this project:
Backend Developer
Backend DeveloperContributions:6 commits, 12 PRs, 11 comments in 11 months
Contributions summary:Koustav primarily contributed to the `cunn` library, focusing on adding padding options to existing volumetric convolution and max pooling layers. They implemented these padding features in the CUDA kernels, improving the functionality and flexibility of these layers. The user also added the CUDA version of VolumetricMaxUnpooling and improved the code related to SpatialDilatedMaxPooling. This work enhances the library's capabilities for 3D convolutional operations commonly used in deep learning applications.
torch/nn
Nov 2015 - Aug 2016
Role in this project:Backend Developer
Backend DeveloperContributions:10 commits, 20 PRs, 50 comments in 9 months
Contributions summary:Koustav's contributions primarily revolve around enhancing and expanding the functionality of the `VolumetricConvolution` module. Their work includes implementing padding support, optimizing performance by introducing a new module `VolumetricConvolutionMM` for matrix multiplication, and adding a `VolumetricMaxUnpooling` module for use after max pooling operations. They also updated existing modules and added tests to ensure the correct operation of the new and modified features.
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.