CLEAR: Cooperative Latent Flow for Visual Counterfactual Explanations with Accurate Region Preservation
Visual counterfactual explanations (VCEs) should alter a classifier's prediction while remaining realistic and preserving decision-irrelevant content.
ProofPaper ↗
Key points
- Image-space methods provide high-quality counterfactuals but are computationally costly, whereas latent-space methods are more efficient but often less faithful.
- We introduce CLEAR, a latent Flow Matching framework for VCEs.
- CLEAR uses Flow Matching's explicit instantaneous velocity as a reference for the generative direction to decompose classifier guidance into cooperative and conflicting components.
- We further analyze conflict frequency and strength throughout generation to reveal how classifier guidance interacts with the generative process.
Sources (1)
- [1]CLEAR: Cooperative Latent Flow for Visual Counterfactual Explanations with Accurate Region PreservationarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 16, 03:01 PM
Visual counterfactual explanations (VCEs) should alter a classifier's prediction while remaining realistic and preserving decision-irrelevant content.
Image-space methods provide high-quality counterfactuals but are computationally costly, whereas latent-space methods are more efficient but often less faithful.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 16, 2026NestedVLA: Learning to Consolidate and Generate Skills for Vision-Language-Action Models
- Oct 16, 2026ReLaDrift: Recurrent Language Generation via Drifting for Test-Time Scaling
- Oct 16, 2026TriFold: Folding irregular 3D Gaussians into structured latents for 3D Scene Generation
- Oct 16, 2026Learning to Be Guided: Shaping Latent Geometry for Classifier-Free Guidance
- Oct 15, 2026On the SPOT: Enhancing Speculative Decoding for Diffusion Language Models via Commit-Order Prediction
- Oct 15, 2026SUPRA-TEX: 3D Generation with Super-Aligned Texture