Code

Siamese Network s43032053#167

Code

Open

richardchantra wants to merge 32 commits intoshakes76:topic-recognitionshakes76/PatternAnalysis-2024:topic-recognitionfrom richardchantra:topic-recognitionrichardchantra/PatternAnalysis-2024:topic-recognitionCopy head branch name to clipboard

+1,220

Commits

Commits on Oct 1, 2024

• Added GFNet Alzheimer's classification folder and README

  richardchantra

  committed
  63800d3

Commits on Oct 18, 2024

• Changed project to siim-isic_melanoma_siamese

  richardchantra

  committed
  93a1094

Commits on Oct 21, 2024

• Initial preprocessing data pipeline with metadata and train-image folder

  richardchantra

  committed
  0fbc881

• Updated README.MD with report structure. Filled out the problem section

  richardchantra

  committed
  d24a528

• Added Siamese Network with Contrastive Loss and training loop to train.py

  richardchantra

  committed
  0dd3fd1

Commits on Oct 22, 2024

• Added CUDA, checkpoint and logging support to predict.py as well as helper functions to predict in predict.py

  richardchantra

  committed
  b1415d9

• Replaced the convolutional layer with pretrained ResNet50 as a feature extractor

  richardchantra

  committed
  b3eb59b

• Preprocessing of images to support ResNet50 including resizing, convert to tensor and normalizing

  richardchantra

  committed
  db2c954

• Fixed dataloader to pick equal (and random) amount of pairs in dataset.py and enabled pin_memory for GPU data transfer support

  richardchantra

  committed
  ce4fafa

• Added MLP Classifer feeding in the Siamese Network embeddings. Also added docstrings to functions.

  richardchantra

  committed
  55c1d29

• Updated predict.py to use the testset and added more evaluation metrics including accuracy, ROC-AUC and classification report

  richardchantra

  committed
  be71511

Commits on Oct 23, 2024

• Added Siamese Network checkpoint skip position and created a DataManager class to encapsulate data loading and configuration

  richardchantra

  committed
  a132ca5

• Added a function to balance the dataset (undersampling) in DataManager

  richardchantra

  committed
  25a46e2

• Enabled balance_dataset to also perform oversampling

  richardchantra

  committed
  479cd95

• Added more dropout layers and weight decay as regularization to prevent overfitting of embeddings

  richardchantra

  committed
  dbb2b04

Commits on Oct 27, 2024

• Updated the training model parameters like learning rate and weight_decay in siamese network also fixed up image augmentations in dataset.py

  richardchantra

  committed
  c7b4e98

• Added an extra dense layer to the siamese network and updated the batch size to be 256 because there is GPU capacity

  richardchantra

  committed
  0761f88

• Refactored train.py and predict.py by adding components to modules.py

  richardchantra

  committed
  eee1551

• Refactored train.py to perform prediction and evaluation of trained model what was previously performed in predict.py

  richardchantra

  committed
  6c34c6d

• Refactored predict.py to enable prediction of directory of fresh images using saved model

  richardchantra

  committed
  e23c969

• Added header block for each of the 4 files and refactored train.py to use Evaluate from modules.py

  richardchantra

  committed
  3e4020d

• Added function to track of loss for each epoch during siamese training and mlp training

  richardchantra

  committed
  523637b

• Added argparsing functionality for dataset.py for csv_path, img_dir and batch_size

  richardchantra

  committed
  848b1fa

• Added argparsing functionality for train.py for csv_path, img_dir and batch_size, epochs_siamese, epochs_mlp and save_dir for plots

  richardchantra

  committed
  ebaab2b

• Added argparsing functionality for predict.py for image folder_path, siamese_model_path and mlp_model_path

  richardchantra

  committed
  ab40b72

• Added assets and updated README.md

  richardchantra

  committed
  8d15930

• Fixed formatting in README.md

  richardchantra

  committed
  8b4c547

• Fixed formatting in README.md

  richardchantra

  committed
  aa0607c

• Fixed formatting in README.md

  richardchantra

  committed
  f3b7aa0

• Fixed formatting in README.md

  richardchantra

  committed
  23a1778

• Fixed formatting in README.md

  richardchantra

  committed
  345502e

Commits on Nov 17, 2024

• Resolved conflict in README.md

  richardchantra

  committed
  5b6ab4e