A common pipeline

Choose DL architecture

Load pretrained weights from models trained on ImageNet

Label/Annotate some data

Finetune your model

2 Problems

Transfer learning - ImageNet weights

Limited budget - Limited task labels

Solutions

Synthetic data

Domain Adaptation

Active Learning

Label propagation

Weakly supervised learning

Interactive annotation

Human in the loop

Self-supervised Learning

Multimodal Learning

Unsupervised Learning