Research map / Image generation / Autoregressive and latent generation
Counterfactual Generative Networks
Axel Sauer, Andreas Geiger
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 |
|---|---|---|---|
| Train classifiers on counterfactual images and predict the shape mechanism's label | MNIST classifiers · MNIST variants with shape-only test signal | out-of-domain accuracy | Improved |
| Add more counterfactual training examples while keeping real data fixed (ablation) | MNIST classifiers · MNIST test domains | test accuracy | Improved |
| Generate several counterfactuals per sampled noise vector (ablation) | MNIST classifiers · MNIST test domains | test accuracy | Improved |
| Use an even ratio of real and counterfactual images (ablation) | ImageNet classifier · ImageNet | classification performance | Improved |
| Increase the counterfactual-to-real image ratio above 1 (ablation) | ImageNet classifier · ImageNet | classification performance | No clear effect |
| Use a counterfactual-to-real image ratio below 1 (ablation) | ImageNet classifier · ImageNet | classification performance | Hurt |
| Remove the shape loss (ablation) | ImageNet CGN · ImageNet composite-image generation | Inception Score and mask quality | Hurt |
| Add the shape loss, including binary-mask and mask-size constraints (ablation) | ImageNet CGN · ImageNet composite-image generation | Inception Score and mask quality | Improved |
| Remove the texture loss (ablation) | ImageNet CGN · ImageNet mechanism disentanglement | mechanism disentanglement | Hurt |
| Add the texture loss (ablation) | ImageNet CGN · ImageNet mechanism disentanglement | mechanism disentanglement | Improved |
| Remove the background loss (ablation) | ImageNet CGN · ImageNet mechanism disentanglement | mechanism disentanglement | Hurt |
| Add the background loss (ablation) | ImageNet CGN · ImageNet mechanism disentanglement | mechanism disentanglement | Improved |
Closest papers by meaning
- Counterfactuals uncover the modular structure of deep generative models (ICLR 2020 · similarity 0.90)
- Counterfactual Image Editing (ICML 2024 · similarity 0.89)
- Benchmarking Counterfactual Image Generation (NeurIPS 2024 · similarity 0.88)
- Counterfactual Generative Modeling with Variational Causal Inference (ICLR 2025 · similarity 0.87)
- V-CECE: Visual Counterfactual Explanations via Conceptual Edits (NeurIPS 2025 · similarity 0.87)
- Rethinking Visual Counterfactual Explanations Through Region Constraint (ICLR 2025 · similarity 0.86)
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.
