Posted:
Implementation of the paper KinectFusion by Microsoft using C++ and CUDA. Presentation video can be seen below. Normal map for the predicted surface for a frame in TUM RGB-D dataset. Read more...
Posted:
In this project I’ve developed a 3D object detection and tracking pipeline for autonomous driving with Python and ROS without any labeled 3D data. Our hardware were only a Livox mid-100 lidar sensor and a RealSense camera. After calibrating the sensors, the pipeline is as follows: Get 2D instance segmentation of the current camera image using YOLACT. Update the id numbers of the detected entitites using SORT Project the masks into 3D point cloud using LDLS Compute the 3D bounding boxes for the detected areas. Read more...
Posted:
Implemented vanilla Tracktor with Faster-RCNN using PyTorch. Added a basic motion model and also a reidentificaiton network as the paper suggested for Tracktor++ . Developed using the MOT16 benchmark and achieved following results on the training set. My results State of the art Tracktor++ results Source code can be found here Read more...
Posted:
We are extending the paper “Divergence-Free Shape Correspondence by Deformation” and representing a motion sequence as a time dependent vector field. We solve correspondence and matching problems for the whole sequence during the optimization. Some of the current results can be seen below. Blue is shape is our estimated deformation and orange is the ground truth in the Dynamic FAUST dataset. Source code will be made available when the project is completed. Read more...
Posted:
Abstract: Sketches are probabilistic data structures that can provide approximate results within mathematically proven error bounds while using orders of magnitude less memory than traditional approaches. They are tailored for streaming data analysis on architectures even with limited memory such as single-board computers that are widely exploited for IoT and edge computing. Since these devices offer multiple cores, with efficient parallel sketching schemes, they are able to manage high volumes of data streams. Read more...