Explainability in Computer Vision
Explainability in Computer Vision
12 févr. 2026 - 14:00This presentation introduces explainability in computer vision and discusses limitations of post-hoc explanationmethods. An alternative approach based on intrinsic interpretability is then presented, where importance is learned explicitly and integrated into the model’s computation. The presentation concludes with a discussion of limitations of existing evaluation metricsand challenges in intervention-based explainability.