DL_framework

Supported tasks

Classification

• MLP
• Simple CNN
• Torchvision ResNet
• SimpleViT
• MLP-Mixer
• ConvMixer
• Transformer (ECG classification)

Dense Prediction

• Noise2Noise (Gaussian noise)
• Noise2Noise (Text overlay, results not very good)

Generative Model

• Vanilla VAE
• Simplified DDPM (UNet)
• NCSN v2 (The EMA resume may not correctly handled)
• More complexed diffusion models
• GAN

Normalizing flow

• NICE

• MADE

• Neural ODE for MNIST

• Naive Continuous NF

• FFJORD

• ECG_ST_MEM

TO Check

• contrastive loss

Other Components

• wandb support
• RandAugment
• AugMix
• mixup
• linear warmup
• save ckpt by metric (save the best model)

How to run

To start training, run the following command:

./train.sh