Research map / Image generation / GANs
Orthogonal Jacobian Regularization for Unsupervised Disentanglement in Image Generation
Yuxiang Wei, Yupeng Shi, Xiao Liu, Zhilong Ji, Yuan Gao, Zhongqin Wu, Wangmeng Zuo
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 |
|---|---|---|---|
| add OroJaR to GAN training (layers L0~3) (ablation) | simple GAN · Dsprites | VP: 30.9 -> 54.7 | Improved |
| apply OroJaR to multiple layers end-to-end instead of SeFa first-layer post-processing | OroJaR · Dsprites | VP: 48.6 -> 54.7 | Improved |
| replace Hessian Penalty with OroJaR | ProGAN · CLEVR-Complex | PPL: 74.7 -> 10.4 | Improved |
| replace Hessian Penalty with OroJaR | ProGAN · CLEVR-U | VP: 79.2 -> 90.7 | Improved |
| replace Hessian Penalty with OroJaR | ProGAN · Edges+Shoes | VP: 28.6 -> 32.3 | Improved |
| replace Hessian Penalty with OroJaR | simple GAN · Dsprites | VP: 48.5 -> 54.7 | Improved |
| replace Hessian Penalty with OroJaR | ProGAN · CLEVR-Complex | FID: 7.1 -> 10.7 | Hurt |
| replace Hessian Penalty with OroJaR | ProGAN · Edges+Shoes | FID: 17.3 -> 16.1 | Improved |
| extend OroJaR from first layer only (L0) to first four layers (L0~3) (ablation) | simple GAN · Dsprites | VP: 47.6 -> 54.7 | Improved |
| extend OroJaR from L0~3 to L0~4 (one more layer) (ablation) | simple GAN · Dsprites | VP: 54.7 -> 52.3 | Hurt |
| apply OroJaR to a deeper single layer (L4) instead of L0 (ablation) | simple GAN · Dsprites | VP: 47.6 -> 35.0 | Hurt |
| apply OroJaR to multiple layers (L0~2) instead of the single layer L2 (ablation) | simple GAN · Dsprites | VP: 50.2 -> 53.5 | Improved |
Closest papers by meaning
- The Hessian Penalty: A Weak Prior for Unsupervised Disentanglement (ECCV 2020 · similarity 0.93)
- Overcoming the Disentanglement vs Reconstruction Trade-off via Jacobian Supervision (ICLR 2019 · similarity 0.89)
- OOGAN: Disentangling GAN with One-Hot Sampling and Orthogonal Regularization (AAAI 2020 · similarity 0.89)
- Learning Disentangled Representations with Latent Variation Predictability (ECCV 2020 · similarity 0.88)
- Disentangling Factors of Variation with Cycle-Consistent Variational Auto-Encoders (ECCV 2018 · similarity 0.88)
- Disentanglement as Identifiable Pushforward Factorisation (NeurIPS 2026 · similarity 0.85)
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.
