This is the project part of the training course, Deep Learning for Microscopy Image Analysis at Marine Biological Laboratory in 2023. Read more here.
What we want to accomplish learning:
- Basic supervised learning
- Improving generalizability of the model: Domain adaptation, pre-training, and fine-tuning
Proposed plan
- Train a U-Net with your own dataset and assess the performance.
- Use a pre-trained network from other members and fine-tune it for your dataset: we can try all the different combinations. We can try different pre-training and fine-tuning strategies. Then, assess the performance.
- We create a confusion matrix of all these learning conditions and evaluate which method is the best.
Bonus challenge: Investigate how we can train one generalist network that performs almost as well as specialist networks.