Welcom to Duy Khanh channel

This is the modified of Resnet50 model for CIFAR10 dataset.

Model summary (Resnet_Unet_model.py)

Resnet50 + 1 block Unet

Data preparation

Cifar10-dataset transform

• Crop, padding

• Rotate

• Flipping (vertical, horizontal)

• Normalization

Training

• Loss: Cross Entropy

• Optimizer: SGD (lr = 0.001)

• Scheduler: ReduceLROnPlateau(maximize test accuracy)

• If train_acc >99 for 30 epochs continously -> Repeat transform with different ratio

Checkpoint file link

"https://drive.google.com/file/d/1kY6iR6XI03o6DWUfqf9s2KFh1Z4cw02c/view?usp=sharing"

Result

Loss and accuracy Confusion matrix