rodekruis / caladrius Goto Github PK

• Change Output Layer Size
• Switch to Regression Loss
• Design Threshold System
• Update Evaluation Methods

Do I need to save the pictures before and after the disaster in the original data set into the Before and After folders respectively?
Are the labels of the pictures before and after the disaster stored in a labels file?

Current implementation applies random transformations to before and after images.

The same transformation should be applied on before and after images.

For debugging any code refactors or new features not directly related to training the neural network, it would be useful to be able to execute run.py on a not-so-powerful laptop.

tolist returns a single item and not a list if the tensor contains only one item.

App throws error variable must be iterable on trainer.py

It seems that pydoc works best if the docstrings are contained in an installed package

Refactor the report interface into React components

Setup a changelog file and start numbering versions

Get a standard satellite dataset so the model can be used globally

Setup a Docker image for the code

Show a heatmap layer In leaflet map widget.
Button to make the map full screen.

Information about administrative regions need to be added to the building coordinates file, so that the final report can group the damage assessment by region.

Some of the building images have large regions of empty pixels. In the case of the drone data, we should check if any of the other files provide better coverage, otherwise discard the building.

To create a useful damage report, the address that corresponds to each building shape should be included in the main shape file.