Antoreep Jana

Single-handedly wrote scripts to automatically quantize the latest TF and torch popular models and deploy to the company software. Quantized popular models in CV domain like DenseNet, InceptionV4, UNet++, U2Net, SegNet, SCI, and many others & deployed on company software.

Lead projects on the CV front for image data manipulation like GANs for Image Enhancement, Similar Image Generation. Created Segmentation pipeline on satellite images of Vegetation Cover, Solar Panels with 91% dice score. Object Detection pipelines to detect buildings, cars in satellite images with 90% mAP. Worked on various spectra of Images like SAR, LIDAR etc.

Operated single-handedly graph neural networks achieving 85% accuracy score and made modularized python scripts. Implemented object detection Pipelines using yolov4 integrated with Weights & Biases achieving performance score 94% [email protected]. Acquired high-quality datasets as well and maintained the training pipeline of 25K images.

Yielded data from MongoDB servers under 5 seconds through help of optimized python scripts. Engineered optimized data analysis to fetch customer behavioural pattern analysis to be utilized for A/B analysis resulting in 20% growth in retention. Planned Computer Vision training pipeline along with dataset preparation. Designed models with accuracy score of 96% to auto-tag the user posts.

First Class First Division