Vivaswat Suresh

Hi! I’m a computer vision engineer at DeepWater Exploration (first one woo!!) who’s passionate about computer vision and GPU acceleration. I’ve worked on SLAM, post-processed 3D reconstruction, various types of stereo depth, underwater color correction, segmentation, object detection, and LOADS more computer vision problems (most of which run in real-time in DWE’s C++/CUDA SDK). Besides just the classic Pytorch and python ML stuff, I love getting my hands dirty with low level C and CUDA code, and my favorite thing is to rethink problems with highly parallel solutions in mind. My next big personal project is seeing whether or not I can take a ViT-B/16 backbone and accelerate it by writing the CUDA and CUTLASS optimizations by hand. My goal is to beat TensorRT with fp16 compilation!! Keep a lookout!