Skip to content

Latest commit

 

History

History

trainings

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
GANs
GANs
 
 
convnext
convnext
 
 
resnetv2
resnetv2
 
 
visionTransformers
visionTransformers
 
 
xception
xception
 
 
README.md
README.md
 
 

README.md

Convolutional Neural Network Trainings

Brief summary about the training configurations used:

Transfer Learning Freeze Layers Balance Classes Segmentation Data Augmentation Normalization Custom Optimizer Callback ConvNeXt ResNetV2 Xception
❌ ❌ ❌ ❌ ❌ ❌ ❌ ❌ Notebook Notebook Notebook
❌ ❌ ✔️ ❌ ❌ ❌ ❌ ❌ Notebook Notebook Notebook
✔️ ✔️ ✔️ ❌ ❌ StandardScaler ❌ ❌ Notebook Notebook Notebook**
✔️ ✔️ ❌ ❌ ❌ ❌ ❌ ❌ Notebook Notebook Notebook
❌ ❌ ❌ ✔️ ❌ ❌ ❌ ❌ Notebook Notebook Notebook
✔️ ✔️ ✔️ ✔️ ✔️ ❌ ✔️ ❌ Notebook** Notebook** Notebook**
❌ ❌ ❌ ❌ ✔️ ❌ ❌ ❌ Notebook** Notebook** Notebook**
✔️ ✔️ ❌ ❌ ✔️ ❌ ❌ ❌ Notebook** Notebook** Notebook**
✔️ ✔️ ✔️ ❌ ✔️ ❌ ❌ ❌ Notebook** Notebook** Notebook**
✔️ ✔️ ✔️ ❌ ✔️ MinMaxScaler ✔️ ❌ Notebook** Notebook** Not trained due to low RAM**
✔️ ✔️ ❌ ❌ ✔️ ❌ ✔️ ❌ Notebook** Notebook** Notebook**
✔️ ✔️ ✔️ ❌ ❌ ❌ ❌ ❌ Notebook** Notebook** Notebook**
✔️ ✔️ ✔️ ❌ ❌ ❌ ✔️ ❌ Notebook** Notebook** Notebook**
✔️ ✔️ ✔️ ✔️ ❌ ❌ ✔️ ❌ Notebook** Notebook** Notebook**

Obs: Execution in Google Colaboratory with High-RAM (25.5 GB) and T4 as GPU.

** Execution in Google Colaboratory with High-RAM (83.5 GB) and A100 as GPU.

Vision Transformer Training

Model ViT_B/32 and TPU. Batch size was equal for all runs: 8 for each TPU core (64). Worth to notice that classes are unbalanced as well as the others parameters weren't calculated.

Epochs Image Resolution Accuracy without fine-tuning Accuracy with fine-tuning Notebook
100 224 0.02929688 0.7998047 Notebook
10 224 0.03125 0.7265625 Notebook
10 256 0.02539062 0.7158203 Notebook

Obs: Execution in Google Colaboratory with Standard RAM (12.7 GB) and TPU.

Generative Adversarial Networks (GANs) Training

Brief summary about training configurations:

Transfer Learning Freeze Layers Balance Classes Segmentation Data Augmentation GAN model Normalization Custom Optimizer ConvNeXt ResNetV2 Xception
✔️ ✔️ ❌ ❌ ✔️ 4 ❌ ❌ Notebook Notebook Notebook
✔️ ✔️ ❌ ❌ ✔️ 4 StandardScaler ❌ Notebook Notebook Notebook
✔️ ✔️ ❌ ❌ ✔️ 4 MinMaxScaler ❌ Notebook Notebook Notebook
✔️ ✔️ ❌ ❌ ✔️ 3 ❌ ❌ Notebook Notebook Notebook
✔️ ✔️ ❌ ❌ ✔️ 3 StandardScaler ❌ Notebook Notebook Notebook
✔️ ✔️ ❌ ❌ ✔️ 3 MinMaxScaler ❌ Notebook Notebook Notebook
✔️ ✔️ ❌ ✔️ ✔️ 3 MinMaxScaler ❌ Notebook Notebook Notebook
✔️ ✔️ ❌ ✔️ ✔️ 3 ❌ ❌ Notebook Notebook Notebook

Execution in Google Colaboratory with High-RAM (83.5 GB) and A100 as GPU.