Class Activation Map (CAM and Grad-CAM) Analysis of fine-tuned CNNs with transfer learning for Pokemon classification task to understand the features learned by deep CNN

Computer tomography (CT) scans are one of the only ways to diagnose lung diseases. With the rise of lungrelated issues in recent times, it would be a boon if a system existed that would aid medical professionals to diagnose diseases from chest CT scans. This paper proposes to develop such a system in the form of two separate modules using basic …

Class activation maps for high risk and low risk patients in lung adenocarcinoma

High Resolutions Class Activation Maps usage script. Implementation from HR-CAM paper.

Detecting Severe Malaria Anaemia and investigating the morphological characteristics of red blood cells at its presenc

Computer Vision with PyTorch for Medical Image Analysis

A collection of my Jupyter notebooks, showcasing my exploration and learning journey in the field of Computer Vision

Generates class activation maps for CNN's with Global Average Pooling Layer Keras

Some Class Activation Map methods implemented in Pytorch for CNNs

MVA Master School Project - Weakly Supervised Semantic Segmentation

This work study the "Activation/Saliency Map" in image classification, which emphasize the regions in a image where model focus on to give the final predication result.

Special Project - CA classification (2019 Fall)

A Class Activation Map is a weighted activation map generated for each image, helping us to identify the region a CNN is looking at while classifying an image.

DeviceScope: An Interactive App to Detect and Localize Appliance Patterns in Electrical Consumption Time Series

Repository containing code to run Score-CAM algorithm available on https://arxiv.org/pdf/1910.01279v1.pdf.

Lightweight Neural Network for Semantic Segmentation using Knowledge Distillation (Accepted by AICAS 2022)

Tutorial to show how to extract object localization

saliency map, adversarial image, (gradient) class activation map

In this project we use a Lightweight-CNN based model to classify instruments from the Freesound audio data set. We make use of Mel-Spectrogram features from the input audio data as the input to the CNN model. To add robustness to the model, we use a novel data augmentation technique based on the Cut-Mix algorithm.