Olivier Laurent

:sparkles: Finalize Depth Estimation, add DeeplabV3, & Add Selective Classification

1

**Depth Estimation** - [x] Add DepthEstimationRoutine - [x] Add MUADDepth Datamodule - [x] Add Metrics - [x] Add iRMSE and iMAE (solve the conventions of the applications of the metrics)...

:sparkles: Add LPBNN, Adaptive ECE, start supporting Depth estimation & Improve segmentation

1

To do: - [ ] Merge #88 - [ ] Update documentation - [ ] Fix coverage - [ ] Add everything to Paperswithcode Closes #83 & #89

Currently, validation may fail even for relatively small models such as DeepLabV3 when using large image crops. Calibration Error is the worst metric but others seem to also have huge...

Original code: https://github.com/giannifranchi/LP_BNN Original paper: https://arxiv.org/abs/2012.02818

Paper: https://arxiv.org/abs/2108.00968 Implementation: https://github.com/giannifranchi/deeplabv3-superpixelmix More particularly, the transform is here: https://github.com/giannifranchi/deeplabv3-superpixelmix/blob/master/datasets/cityscapes_mix.py#L13

- [ ] Add a new routine - [ ] Add a model - [ ] Add NYUv2 - [ ] Add MUAD Depth - [ ] Add metrics

This metric could be added to the default group in the classification routine.

Display the distance with the optimal calibration bins. For instance, with dashed red histograms.

:bug: Fix the learning rate scale

1

Thanks again to @hanruisong00 for the remark. We have to discuss the best way to fix this scaling problem. The current solution is unsatisfactory as it scales all ensemble methods;...

:shirt: Fix plot size in the docs

1

See https://torch-uncertainty.github.io/auto_tutorials/tutorial_scaler.html#sphx-glr-auto-tutorials-tutorial-scaler-py. This goes for all tutorials.