Experience

• 

  Ieee nitk

  Sept 2019 - Jan 2022

  1. Conducted and supervised Impulse 2022, an online signal processing hackathon involving online workshops, talks and a hackathon.2. Managed and mentored the signal processing based executive projects of the Diode SIG. Mentored and led a team of junior executive members in a project titled "Hiding Images Inside Images".

  • SPS Chair

    Apr 2021 - Jan 2022

  • Executive Member

    Sept 2019 - Apr 2021
• 

  Indian institute of science (iisc)

  May 2021 - Aug 2021

  Research intern

  I worked under the guidance of Dr. Chandra Sekhar in a project that aimed to detect various diseases and abnormalities using Deep Learning in Retinal Fundus Images.

• 

  Technische universität braunschweig

  May 2021 - Oct 2021

  Research intern

  1. Developed an efficient codebase for training and testing for semantic segmentation of sclera regions in the eye images. 2. Acquired partial annotations using a game where the partial masks are saved as players competed for scoring regions.3. Acquired an F1 score of 0.94 on the test segmentation set using multiple partial annotations.

• 

  Sixsense

  Jan 2022 - Apr 2022

  Computer vision intern

  1. Worked on detecting and classifying defects in semiconductor chips using Faster RCNN.2. Trained a stochastic automatic augmentation framework based on Fast AutoAugment on a ResNet50 model to techniques for several public datasets like CIFAR-100 and in-house datasets which improved the average accuracy by 2.3%. 3. Integrated the automatic augmentation to the defect detection pipeline, improving the accuracy by 1.4%.

• 

  Department of radiology, university of michigan

  Aug 2022 - Apr 2023

  Research assistant

  1. Trained an attention-UNET-based model for aortic segmentation, enhancing the accuracy and efficiency of the Vascular Deformation Mapping pipeline, resulting in an improvement of 3% in the F1-score, particularly around aortic walls.2. Implemented corrections to an Elastix-based CT Registration Pipeline, improving the elastic registration performance of the pipeline for large deformations in the aortic walls. The corrections resulted in the detection of tissue growth by an improved recall of 8%. Show less

• 

  Electrical and computer engineering at the university of michigan

  May 2023 - now

  Research associate at kim's lab

  1. Currently working on optimizing a LiDAR+Camera fusion based 3D object detection pipeline using generative models for Bird-Eye-View feature completion. 2. Finetuned the state of the art BEVFusion for 3D bounding box detection by only utilizing 50% of the lidar beams, achieving a mAP of 0.601 and NDS of 0.63 on NuScenes dataset using subsampled point clouds. 3. Developed novel diffusion sampling algorithms for inverse imaging problems, enhancing the quality and fidelity of the samples of latent diffusion models, achieving an FID score of 37.2, an improvement of 17.2% over the baseline model. Show less

• 

  Skylark labs

  Jun 2023 - Aug 2023

  Machine learning intern

  1. Designed a framework using a pre-trained RegNet model to achieve a recall of 65% in self-learning new categories by storing multi-scale quantized features to recognize pre-trained classes.2. Trained a vector-quantized feature extractor to learn efficient multi-scale features of objects in natural scenes, enhancing the accuracy of the model by 12% to detect objects from newly learned classes.3. Implemented the system to work on a single core of a CPU to run at about 3 fps while storing features of new classes encountered. Show less

• 

  University of michigan robotics department

  Aug 2023 - Dec 2023

  Graduate student instructor

  Instructor: Jason CorsoCourse: EECS 504: Foundations of Computer VisionConducted discussion sessions on research papers and programming assignments involving Image Segmentation, Feature Tracking, Motion, and Stereo.

• 

  Computer science and engineering at the university of michigan

  Jan 2024 - Apr 2024

  Graduate student instructor

  Instructor: Jeong Joon ParkCourse: EECS 442: Computer Vision