Transfer Learning vs Fine-Tuning

Transfer learning

Transfer learning means taking a pretrained model (SAM3) and adapting it to a new domain or task instead of training from scratch.

In this project:

• Start from pretrained SAM3 weights.
• Train only selected parts (or small adapter modules) on forest data.

Fine-tuning

Fine-tuning is the training step used during transfer learning.

In this project:

• Continue training SAM3 on labeled forest imagery.
• Use lower learning rates and domain-specific data.

Relationship:

• Transfer learning is the strategy.
• Fine-tuning is the mechanism.

See also:

• Phased Strategy
• Transfer Learning Methods