Code used in Gradient-Based Quantification of Epistemic Uncertainty for Deep Object Detectors
August 16, 2024 ยท View on GitHub
We deliver all code used for our experimental setups and for the generation of the shown plots. The folders yolov3-torch, faster-rcnn-torch and retinanet-torch contain the object detection pipelines. We are unable to provide weight files for the networks due to upload constraints. The object detectors all implement a pipeline for computing gradient uncertainty metrics, a more detailed structure can be found in the respective README.md-files included in the folders.
The folder uncertainty_aggregation contains the framework we used for aggregating uncertainty metrics in meta classification and meta regression, as well as the main experimental setups once gradient-based and Monte-Carlo dropout-based uncertainty metrics have been produced. We included a more detailed description in the README.md-file in the folder itself.