Aniket Dash

Some of my work

I develop data-driven tools and machine learning models to extract insights from satellite imagery and audio signals, enabling a deeper understanding of our environment.

I also enjoy experimenting with audio and visual data, often creating intuitive interfaces that bridge complex models with real-world usability.

I mostly work with:

Recent Projects

ChangeSense

Building Instance Segmentation and Reporting

ChangeSense is an instance segmentation model designed to detect and report changes in buildings or manmade structures in satellite imagery over time.

[Code] [Paper]

  • Fine-tuned YOLOv8
  • Flan-t5 model for natural language reporting
CMSN sample 1 CMSN sample 2

Denoisy

Independent Research · 2024

Denoisy is a low-light image restoration pipeline using hybrid frequency–spatial priors. It leverages FFT-based attention to enhance fine-grained image textures in extreme illumination settings.

[Code] [Paper]

  • Optimized for real-time enhancement on edge GPUs
  • Outperforms U-Net baseline by +8% PSNR on night scenes
Denoisy output 1 Denoisy output 2

CMSN: Cloud Removal Network

Inspired by 2022 ASOC Paper · Remote Sensing

A multi-temporal synchronous network designed for reconstructing cloud-free Landsat imagery using coarse-to-refined fusion.

[Code] [Paper]

  • 30% improvement in temporal consistency metrics
  • Integrates ResFFTConv, SCM, and MFAM blocks
CMSN sample 1 CMSN sample 2 CMSN sample 3

Agri-scene Classifier with SimCLR

simCLR

Identifying agriculture zones using contrastive learning on satellite images.

[Code] [Paper]

  • Used ResNet18 and ResNet50 for backbone
  • Gauging NDVI and classifying scenes
SimCLR sample 1 SimCLR sample 2 SimCLR sample 3 SimCLR sample 4 SimCLR sample 5

EduSync

Application for Instructors

AI-powered dashboard to manage classes, track progress, and generate reports.

[Code]

EduSync sample 1 EduSync sample 2 EduSync sample 3 EduSync sample 4 EduSync sample 5

Indictts-deepfake detector

Wav2vec2

AI-generated speech detector across 16 Indian languages.

[Code]

Low light Depth Estimator

UNET

Depth Estimation: Generating depth maps from challenging low-light captures.

[Code]

Fake News Detector

Random Forest

Fake News Detection (Malayalam): Detecting misinformation using custom neural networks and classic ML models.

[Code]

Business Data Management

Pandas | Excel

Shawarma Truck Data Project: Improving business decisions for a local food truck using data insights.

[Code]

Juke

Music Streaming Application

A music streaming app built for local creators and listeners with access control.

[Code]

Chime

Blog

A multi-user blog platform to share thoughts, follow users, and engage.

[Code]