-
[2024.12.31] We released a baseline for interpretable object detection methods based on mmdetetion version 3.0, supporting GradCAM, ODAM, and D-RISE algorithms. Please see version3
-
[2024.11.25] Our paper Interpreting Object-level Foundation Models via Visual Precision Search is now visible on https://arxiv.org/abs/2411.16198, and the code will be released at https://github.com/RuoyuChen10/VPS after publish.
-
[2023.01.17] We released a baseline for interpretable object detection methods based on mmdetetion version 3.0, supporting GradCAM, and D-RISE algorithms. Please see version2.25.2
We have conducted theoretical and experimental analyses demonstrating that VPS is a highly faithfulness interpretation method for explaining object-level model decision (object detection or visual grounding), applicable to both multimodal object-level foundation models and traditional target detection models. The paper is available at https://arxiv.org/abs/2411.16198, and the code will be released at https://github.com/RuoyuChen10/VPS following acceptance.
