Akshansh Mishra

Akshansh MishraAkshansh MishraAkshansh Mishra

Akshansh Mishra

Akshansh MishraAkshansh MishraAkshansh Mishra
  • Home
  • Publications
  • Ongoing Research
  • Masters Thesis
  • Atomistic Modeling
  • Machine Learning
  • Music and Algorithms
  • AUGMENTED REALITY
  • Additive Manufacturing
  • Architected Metamaterials
  • Composite Materials
  • Developed APPS
  • STL Gallery
  • Altro
    • Home
    • Publications
    • Ongoing Research
    • Masters Thesis
    • Atomistic Modeling
    • Machine Learning
    • Music and Algorithms
    • AUGMENTED REALITY
    • Additive Manufacturing
    • Architected Metamaterials
    • Composite Materials
    • Developed APPS
    • STL Gallery
  • Home
  • Publications
  • Ongoing Research
  • Masters Thesis
  • Atomistic Modeling
  • Machine Learning
  • Music and Algorithms
  • AUGMENTED REALITY
  • Additive Manufacturing
  • Architected Metamaterials
  • Composite Materials
  • Developed APPS
  • STL Gallery

Music and Algorithms

Working of class activation maps

 

When a neural network looks at a video, it doesn't process the entire frame equally. Class Activation Maps reveal exactly where the model is focusing its attention. Think of it as highlighting the parts of an image that matter most for recognition. Let's take an example of a classic scene from season 3 of Stranger Things (Chapter 8: The Battle of Starcourt). 



Algorithms for precise motion analysis

Real-time motion analysis system combining MediaPipe pose estimation and YOLO object detection. Track multiple people simultaneously with color-coded skeletons, calculate movement velocity and joint angles, analyze energy levels, and generate detailed performance metrics. Handles crowded scenes while providing persistent ID tracking, individual movement data, and comparative rankings with automatic validation. 


GitHub Repo:  https://github.com/akshansh11/Motion-Movement-Tracker 

Real-time music visualization

A Python-based music visualization tool that creates stunning, beat-synchronized visual representations from video files containing audio. Perfect for visualizing speaker outputs, music performances, or any audio content. 


GitHub Repo:  https://github.com/akshansh11/Audio-Video-Visualizer/tree/main

Deep Learning models to visualize the video embeddings

Video embeddings transform high-dimensional visual data into compact numerical representations that capture semantic meaning, color patterns, and motion dynamics. These embeddings are projected into 2D space using Principal Component Analysis, we can visualize how different aspects of a video are encoded and clustered over time. Let's understand this by taking an example of a scene from one of the classic anime series (Beyblade V Force: Episode 45).

Multimodal Machine Learning based Cellular Structures Design

 The generation mechanism operates through a multi-stage pipeline: First, the input MP4 video is uploaded to multimodal language API, which employs transformer-based multimodal learning to simultaneously process visual and temporal information. The AI model analyzes frame-by-frame content to identify geometric primitives, motion vectors, color gradients, and spatial patterns. 

Copyright © 2025 Akshansh Mishra - Tutti i diritti riservati.


Gestito da

Questo sito web usa cookie.

Utilizziamo i cookie per analizzare il traffico sul sito web e ottimizzare l'esperienza con il tuo sito. Accettando l'uso dei cookie da parte nostra, i tuoi dati saranno aggregati con i dati di tutti gli altri utenti.

Accetta