Research map / Probabilistic and scientific modeling / Causal discovery
Interpreting Low-Level Vision Models With Causal Effect Maps
Jinfan Hu, Jinjin Gu, Shiyao Yu, Fanghua Yu, Zheyuan Li, Zhiyuan You, Chaochao Lu, Chao Dong
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use Flickr2K instead of DIV2K as natural image library (ablation) | CEM · SR, SwinIR | positive effect patches (%): 22.85 -> 23.24 | No clear effect |
| use Flickr2K instead of DIV2K as natural image library (ablation) | CEM · SR, RCAN | positive effect patches (%): 31.07 -> 30.85 | No clear effect |
| use Flickr2K instead of DIV2K as natural image library (ablation) | CEM · SR, SRResNet | positive effect patches (%): 7.04 -> 7.05 | No clear effect |
| use density-based sampling instead of uniform sampling in accelerated CEM (ablation) | CEM C3F50 · SR | similarity score (%): 77.72 -> 82.62 | Improved |
| use 3 coarse interventions instead of 1 (ablation) | CEM C1F50 · SR | similarity score (%): 75.86 -> 82.62 | Improved |
| increase fine-stage interventions from 10 to 50 (ablation) | CEM · SR | similarity score (%): 65,32 -> 82.62 | Improved |
| increase fine-stage interventions from 30 to 50 (ablation) | CEM · SR | similarity score (%): 78.68 -> 82.62 | Improved |
| use transformer architecture (X-Restormer) vs SRCNN for SR | SR models · SR | positive effect patches (%): 3.06 -> 38.27 | Improved |
| compare denoising to super-resolution with same architecture (SwinIR) | SwinIR · DN vs SR | positive effect patches (%): 22.85 -> 4.85 | Hurt |
| take intervention patches from the input image itself instead of a natural image library (ablation) | CEM · SR | CEM map | Hurt |
Closest papers by meaning
- Causal Concept Graph Models: Beyond Causal Opacity in Deep Learning (ICLR 2025 · similarity 0.86)
- Causal Tracing of Object Representations in Large Vision Language Models: Mechanistic Interpretability and Hallucination Mitigation (AAAI 2026 · similarity 0.85)
- Learning 1D Causal Visual Representation with De-focus Attention Networks (NeurIPS 2024 · similarity 0.85)
- The Non-Linear Representation Dilemma: Is Causal Abstraction Enough for Mechanistic Interpretability? (NeurIPS 2025 · similarity 0.85)
- Flexible Context-Driven Sensory Processing in Dynamical Vision Models (NeurIPS 2024 · similarity 0.84)
- From Prompts to Tokens: Internalizing Causal Supervision in Vision-Language Model for Multi-Image Causal Reasoning (ICML 2026 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
