pytorch_etude

The repository is for deep learning example codes using pytorch, pytorch_lighting, and tensorboard.

Examples log in the tensorboard. Some examples provide the script file to demostrate the result of model.

Ex1 Linear Model

In this example, Linear model guesses the slope and bias of the noisy input.

Ex2 Linear Model for MNIST

The model of ex2 consists of linear modules. This model classifies the handwritten image of MNIST.

Plot Good Result	Plot Bad Result	Confusion Matrix	Vector visualization

Ex3 CNN Model for MNIST

Ex3 has higher accuracy than Ex2.

Output of 1st Conv2D	Output of 2nd Conv2D

EX4 Denoise Autoencoder for Denoising

The model of ex4 consists of convolutional networks. This model removes noise from the mnist images to which noise has been added.

Traninig process in Tensorboard	Denoising Result

EX5 CAM(Class Activation Map) for Mnist

This model use GAP(Global Average Pool) to get CAM.

CAM of good result	CAM of bad result

EX6 GradCAM(Class Activation Map) for Mnist

This model use GradCAM to get CAM. (GradCAM code from 1Konny)

GradCAM of good result	GradCAM of bad result

EX7 Interpretation model using CAPTUM

This example interpret the model using CAPTUM.

EX8 R-CNN Model for kaggle bus truck data

This model use R-CNN to detect and classify Kaggle bus-truck data. To reduce complexity, this model use a linear model to classify items instead of SVM in the original paper.

Selective Search Result

RCNN Result

EX9 Fast R-CNN Model for kaggle bus truck data

This model contains RoiPooling to implement fast RCNN.

Fast RCNN Result

EX10 Faster R-CNN Model for kaggle bus truck data

This model use the faster rcnn in torchvision.

Fast RCNN Result

Reference

[1] Modern Computer Vision with PyTorch: Explore deep learning concepts and implement over 50 real-world image applications.