Thijs #2
Thijs #2
Conversation
There was a problem hiding this comment.
Pull Request Overview
This pull request implements multiple enhancements for the XAI system including improved visualization for heatmaps and edges, refactoring model prediction and feature extraction logic, and additional utility functions and tests for robustness.
- Introduced new visualization functions (heatmap, difference maps, side-by-side displays, and edge-enhanced overlays).
- Refactored model.py to replace OpenCV edge detection with scikit-image functions and standardized image preprocessing.
- Added new utility modules for metrics, configuration, and file operations; improved test coverage across the codebase.
Reviewed Changes
Copilot reviewed 52 out of 53 changed files in this pull request and generated 2 comments.
Show a summary per file
| File | Description |
|---|---|
| visualization/heatmap.py | New heatmap and multi-image plotting functions |
| visualization/init.py | Initialization file for visualization functions |
| utils/metrics.py | Metrics calculations and interpretation functions |
| utils/file_utils.py | Directory and file path utility functions |
| utils/config.py | Configuration parameters for the XAI system |
| tests/* | Extensive tests for model functionality and cv utilities |
| model.py | Refactored model integration and feature extraction |
| edge_detection/* | New functions for edge detection and overlay visualization |
Files not reviewed (1)
- results/summary/0_Edinburgh_Nat_Gallery.jpg_nmasks_100_pkeep_0.5_res_8_integrated_metrics.npz: Language not supported
| probabilities[0] -= 0.5 | ||
|
|
||
| if mean_diff > 400: | ||
| mean_diff = 400 | ||
| probabilities[0] = probabilities[0] - 0.3 | ||
| probabilities[0] -= 0.5 |
There was a problem hiding this comment.
The adjustment logic subtracts 0.5 from the probability when mean_diff is less than 99, which may result in negative probability values. Please verify and consider adding a guard to prevent probabilities from dropping below 0.
| probabilities[0] -= 0.5 | ||
|
|
||
| if mean_diff > 400: | ||
| mean_diff = 400 | ||
| probabilities[0] = probabilities[0] - 0.3 | ||
| probabilities[0] -= 0.5 |
There was a problem hiding this comment.
Subtracting 0.5 from the predicted probability in multiple conditions might lead to negative outcomes. Ensure that the adjustment algorithm is correctly calibrated and that a lower bound is enforced.
This pull request addresses the initial broadcast issue with using DIANNA ( issue #1) by making use of a new custom RISE function. Key changes include a new visualization overlay function, and improved preprocessing and feature comparison logic.
Visualization Enhancements:
visualize_edge_heatmap_overlayinedge_detection/visualizer.pyto overlay edge maps and heatmaps for improved interpretability. This function includes configurable parameters for edge detection, transparency, and color mapping. (edge_detection/visualizer.py)Model Refactoring:
model.pyto replace OpenCV-based edge detection withskimagefunctions, integrate a standardizedpreprocess_imagefunction, and improve batch processing with progress indicators. (model.py) [1] [2]compare_image_with_datasetto use the refactored edge detection and feature extraction logic, including adjustments to probability calculations and output formatting. (model.py)Additional Changes: