Research & Engineering Collaborator
Current- Performed finer-grained classification of offshore marine platforms in Satlas using a pipeline consisting of a novel Swin Transformer backbone, feature pyramid network intermediate, and a classification head.
- Derived finer-grained classes for offshore marine platforms by intersecting NOAA ENC data with Satlas data.
- Ran experiments on Beaker with different low-res training image sizes and data augmentations, creating confusion matrices to analyze models’ performance. Result was deploying a model having 97.3% validation set accuracy.