d_learning

To update GIT:

git push https://github.com/ANTZ314/d_learning.git --force

anomaly:

AE

• Simple reconstructive AutoEncoder
• Reconstructive AutoEncoder for denoising demo
• Reconstructive AutoEncoder for Anomaly Detection score

mvtec_ad

• train.py
• finetune.py
• test.py
• stored_models
  • 2 trained models with loss plots
  • [21-02-2022] model trained on kaggle leather dataset

novelty_det1

• test.py - CHECK IF TESTED??

novelty_det2

• UNTESTED

transfer:

kaggle:

Description:

• class6.py - Modified ResNet-50 model from 100 classes to 6 classes, corresponding to the 6 Kaggle leather dataset catagories
• class6a.py - Above code split and modified for direct testing/classification + Training evaluation
• class6b.py - Custom model created & tested with stored pre-processed dataset + test prediction?

transfer

Description:

• class_4.py
• class_5.py
• class_10.py
• custom1.py
• custom2.py
• inception.py
• resnet50.py
• vgg16.py

ViT

Description: Various ViT code examples tested up to classification (at least)

• ViT_0 - UNTESTED
• ViT_1
  • Classifier tested in Locally & in Colab
  • ViT_1a.py - Original Vision Transformer example from Colab (link in code description)
  • ViT_1b.py - Removed visualisations and other superfluous code
  • ilsvrc2012... - ImageNet Labels for class prediction
• ViT_2 - Classifier tested in Colab
• ViT_3
  • Tested in Colab - Not with custom test image (only Cifar-10)
• ViT_4
  • Good test metrics at the end (Activation Map / Confusion Matrix)
  • Testing incomplete (Colab / Custom Image)