Fine-tuning GBIF models with trap data

[adityajain07]

Experiments to quantify the improvement in accuracy when AMI-GBIF models fine-tuned with AMI-Traps data

[adityajain07]

Experiments (model training):

1. Only AMI-GBIF
2. Ony AMI-Traps
3. AMI-GBIF fine-tuned with AMI-Traps
4. AMI-GBIF + AMI-Traps combined

To do (in-order):

• Solve [Subtask] Issue with webdataset files #21
• Datasplit of AMI-Traps dataset
• Re-organize and test training code. Push it to src. [Subtask] Reporting metadata with the model #23 maybe worked on in this process.
• Get evaluation metrics for the first three experiments
• Combining the AMI-GBIF and AMI-Traps dataset, and performing the 4. experiment

[joseepoirier]

Following discussion between AJ and DR, this task is downprioritised to work on global model first.