geo-deep-learning
geo-deep-learning copied to clipboard
NRCan
Deep learning applied to georeferenced datasets
Removing the part where we are combining the existing GPKG from the post-processing script and create a new python script to do it. This solution will avoid writing on the...
See [define_model()](https://github.com/NRCan/geo-deep-learning/blob/develop/models/model_choice.py#L121). It doesn't seem relevant to add define_model() function to the already existing [define_model_architecture()](https://github.com/NRCan/geo-deep-learning/blob/develop/models/model_choice.py#L17). What define_model() does on top of define_model_architecture() is (1) convert model to a DataParallel model...
The first extractions performed under this image reveal poor performances for WAER and ROAI. Meanwhile the extractions for BUIL and FORE seem to have the usual behavior. Example of artifact:...
Currently, the dataset.bands parameter is a list as [1,2,3], ["R", "G", "B"] or ["red", "green", "blue"]. However, if a model is trained with bands as integers (therefore this param is...
In AOI's \_\_init\_\_, we'd need to allow the use of a layer from ground truth's geopackage that contains a single polygon which defines the extent of interest for ground truth...
Verify_segmentation.py script is meant to be the first step when starting a project with new data. This script should be able to complete (i.e. write it's csv report) no matter...
Not a priority: SegmentationDataset is going to get a huge refactoring in the course of the next weeks while addressing #152, which already started with PR #406 *** Csvs created...
[feature request]: While testing, Pycharm's default working directory cannot always access test data
## Description (Ideally, we'd solve issue #408 before this issue to remove duplicate references to test data) For example, if a test is located under `tests/utils/`, the default working directory...
Multiple tests in our test suite use test data under the `tests/data/`, referenced by different csvs in `tests/tiling/` This test data could be made accessible for all tests via pytest...
At the moment the scheduler option is not call and is written in the old version of hydra. We need to call the scheduler in the training for better results...