PROJECTS
PROJECTS
Embedded Files
Object Detection in Aerial Images
Object Detection in Aerial Images
(Aug. 2019 - Dec. 2019) | (Report)
Aim: Perform object detection in drone images (i.e. VisDrone2019 Dataset).
Development Stages:
EDA - Exploratory Data Analysis.
Train a baseline (RetinaNet) model.
Proposed a novel object detection model for drone images, consisting of U-Net (as feature extractor backbone network), Visual Attention Network (VAN), and Focal loss for classification. Implemented it in PyTorch.
Further experimentation: Experimenting with different variations in the proposed model (such as skip-connected autoencoder for feature extraction, etc.); Hyperparameter-tuning.
Mentor: Prof. Pratik Narang
Compiler Construction
Compiler Construction
(Jan. 2019 - May 2019) | (code)
Developed a small compiler for a custom C-like language.
The compiler consisted of the following modules: lexical analyzer, parser, abstract syntax tree generator, semantic analyzer, type checker, intermediate code generator and machine-level code generator.
Done for partial fulfillment of the course Compiler Construction (CS F363).
Original/Enhanced images using WESPE
Weakly Supervised Photo Enhancer For Digital Cameras
Weakly Supervised Photo Enhancer For Digital Cameras
(April 2019) | (code)
Implemented CVPR-2018 paper titled "Weakly Supervised Photo Enhancer for Digital Cameras" (WESPE) in Keras with TensorFlow Backend.
WESPE translates photos taken by cameras with limited capabilities into DSLR quality photos automatically.
It consists of a novel image-to-image GAN-based architecture that is to be trained under weak supervision (i.e. no need for strong supervision in the form of a large annotated dataset of aligned original/enhanced photo pairs).
Dataset Used: DPED.
Done for partial fulfillment of the course Neural Networks And Fuzzy Logic (BITS F312).
Social Collaborative Filtering by Trust
Social Collaborative Filtering by Trust
(Nov. 2018) | (code)
Implemented a research paper titled "Social Collaborative Filtering by Trust" in PyTorch.
Intuition: Build a high-quality recommender system by tackling the cold-start and data-sparsity problems using additional trust data (i.e follow/friend relationship data) among users collected from social networking sites.
Dataset Used: Epinions.
Done for partial fulfillment of the course Information Retrieval (CS F469).
Convolutional Neural Networks in Scala
Convolutional Neural Networks in Scala
(Nov. 2018) | (code)
Implemented a two-layer deep convolutional neural network in Scala using purely functional programming concepts.
Done for partial fulfillment of the course Principles of Programming Languages (CS F301).
LOGIC PROGRAMMING
LOGIC PROGRAMMING
(Nov. 2017) | (code)
Implemented a logic programming solution in SWI-Prolog to classify a given poem as a haiku, doha, quartet, fusion sonnet, or unknown.
Done for partial fulfillment of the course Logic in Computer Science (CS F214).
Page updated
Google Sites
Report abuse