The cifar-10 CNN classifier with ~92% accuracy. It uses some augmentation of images, contains logging of lots of metrics such as projector, PR-curves, etc.
- Run requirements.bat and then start model.py
- Unpack archives in logs folder
- To start learning, run start_tb.bat and then run model.py
- Before learning delete logs directory, then run model.py
- To see model predictions after learning, run model.py
- DATA directory contains results of my attempt of learning
- Logs directory contains all the results of learning
Epoch Acc and Loss
Epoch LR
Epoch 3
Epoch 84
Automobile Epoch 1
Automobile Epoch 84
Cat Epoch 1
Cat Epoch 84
PCA
UMAP 2D
UMAP 3D
Gradients Beta
Gradients Bias
Main Kernel
Moving Mean
Gradients Gamma
Gradients Kernel
Main Kernel
Moving Mean
Moving Variance
