DeepTreeAttention - PLOS Biology Species Release for all NEON sites

Biosphere - Forest Remote Sensing - Python
Published by bw4sz about 1 year ago

DeepTreeAttention - Submission

This release aligns with the submission of the article. The corresponding cometml experiment is:

cometml experiment

The data for the train/test split is available here:
zenodo record

Biosphere - Forest Remote Sensing - Python
Published by bw4sz over 2 years ago

DeepTreeAttention - HSI + Metadata

This release points to https://www.comet.ml/bw4sz/neontrees/e92e400fbf3c4792b16ef2d77105bba3?experiment-tab=chart&showOutliers=true&smoothing=0&transformY=smoothing&viewId=QkOEZWk4wX8jdSX0nJ3tXT8Xg&xAxis=epoch

• Ensemble Micro F1 is 0.75 and Macro F1 is 0.6
• Cleans test data to remove dead and fallen trees
• Combines height, location information with HSI data.
• No RGB integration

Biosphere - Forest Remote Sensing - Python
Published by bw4sz over 4 years ago

DeepTreeAttention - Tree Model

This is the first release of a hyperspectral tree model.

https://www.comet.ml/bw4sz/neontrees/f72e02543ba14c64b91439126fbe317a?experiment-tab=chart&showOutliers=false&smoothing=0&transformY=smoothing&xAxis=epoch

Biosphere - Forest Remote Sensing - Python
Published by bw4sz over 4 years ago

DeepTreeAttention -

Updates to training layers

• train each subnetwork for less time
• learning rate scheduler is more patient and allows loss to wander
• cleanup of predicted train/test split on resampled ground truth resolution (1m)

Biosphere - Forest Remote Sensing - Python
Published by bw4sz over 4 years ago

DeepTreeAttention - Initial Release

This is the first release of the weak spectral-spatial network model architecture described in Hang et al. 2020. The Houston2018 hyperspectral data were downloaded from the competition website and converted into tfrecords for multi-gpu model training. For more details see the README.md.

Biosphere - Forest Remote Sensing - Python
Published by bw4sz almost 5 years ago