Giving DeepBedMap the bigger picture

[weiji14]

Breaking changes to our DeepBedMap neural network model before the v1.0 release (hopefully soon!). Making better use of recently added datasets such as the Carlson Inlet one (#164), but really it's the new (and still awkwardly sized) 450m resolution MEaSUREs phase-based Ice Velocity data (#165) that we're really trying to better utilize here. Also introducing 'topographic loss' and 'structural loss' to our neural network's cost function and using some Deformable Convolutional layers!

TODO:

• Part 1: Tweaks for new MEaSUREs phase-based Ice Velocity dataset (implement in #165)
  • [x] Increase groundtruth (Y_data) tile size from 8km to 9km (9d79983907f95ff08612746b8ae75b4b61ee176c)
  • [x] ~Use MEaSUREs Ice Velocity's native 450m resolution, i.e. revert a8863e427a4094cdc844a4597d4b19f91d0e6f71~
  • [x] ~Stick to using 450m resolution MEaSUREs Ice Velocity in data_prep.ipynb, revert #65 by not padding input datasets anymore (7cd28af533168f9e4ed0b26211ea84fefba15300)~
  • [x] Re-architect Convolutions in ESRGAN (#78) to use different kernel_size, padding and strides, bilinear resize MEaSUREs Ice Velocity to 500m on the fly (75b74936f11ee011003ef99acdceedb742063bc7)
  • [x] Reverted all the foolish reverts above, while keeping the refactored selective tile function from #167 (d68f2c7dbbcc3e34ff6f8f4350c6fbc50611e1c3)
• Part 2: Further tweaks to Enhanced Super-Resolution Generative Adversarial Network (extends #78)
  • [x] Insert 'topographic loss' into our cost function (d599ee8dd0f428309519253bb2f6e8839f02a2b3)
  • [x] Introduce Structural Similarity (SSIM) Index loss (#172) and adjust loss weightings (59e0122ad97045036c8635e4ec2b3c079c119466)
  • [x] Use Deformable Convolution layers (#171)
  • [x] Comprehensive hyperparameter tuning (5a6c6a663ea78db4e3335f1c72e593514e48a2ab)
  • [x] Create new DeepBedMap DEM model of Antarctica! (db6568e29dc86c578957ab77e4ed5b15306a3e9b, 9708040cb88e4d26646eeca6ed445e039554e6b3)

[review-notebook-app[bot]]

Check out this pull request on ReviewNB: https://app.reviewnb.com/weiji14/deepbedmap/pull/166

You'll be able to see notebook diffs and discuss changes. Powered by ReviewNB.

[ghost]

DeepCode Report (#398375)

DeepCode analyzed this pull request. There are 4 new info reports. 2 warnings and 6 info reports were fixed.