(WIP) Integration of gradcam #feature36

[jasminziegler]

Motivation

As a data scientist, I want to know why a model makes a certain prediction. A heatmap can help me understand where the model is looking at in the input image. This can serve as a sanity check for our models.

Goal

Integrate gradcam code in cgm-rg height workflow and display results in tagging tool.

Implementation

• create height results along with gradcam heatmap
  _generate height results + gradcam in test notebook (like in cgm-rg)
  _use second gradcam implementation from here: https://gist.github.com/RaphaelMeudec/e9a805fa82880876f8d89766f0690b54, to better understand it, read here: https://www.sicara.ai/blog/2019-08-28-interpretability-deep-learning-tensorflow
  _write function that returns name of last conv layer
  _write own function to overlay depthmap and heatmap for debugging

• integrate in cgm-rg
  _re-write function to take on numpy_array only
  _make grad_model global in cgm-rg
  _adapt all relevant files
  _add extract_last_conv_layer_name function
  _add get_height_prediction_and_heatmap_local
  _add/adapt json file in src/workflows
  _complete actual workflow code in src/result_generation/height/height_plaincnn.py

Checklist:

• My code follows the style guidelines of this project
• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests
• New and existing unit tests pass locally with my changes
• My Code is review by 1 People