Research map / Image generation / Autoregressive and latent generation
Pretrained Reversible Generation as Unsupervised Visual Representation Learning
Rongkun Xue, Jinouwen Zhang, Yazhe Niu, Dazhong Shen, Bingqi Ma, Yu Liu, Jing Yang
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 generative pre-training (ablation) | PRG · CIFAR-10 | accuracy: 0.70 -> 0.97 | Improved |
| add generative pre-training (ablation) | PRG · Tiny-ImageNet | accuracy: 0.35 -> 0.71 | Improved |
| add generative pre-training (ablation) | PRG · ImageNet | accuracy: 0.42 -> 0.78 | Improved |
| fine-tune the generative model instead of freezing it (ablation) | PRG · CIFAR-10 | accuracy: 47.10% | Improved |
| fine-tune the generative model instead of freezing it (ablation) | PRG · Tiny-ImageNet | accuracy: 58.04% | Improved |
| use label smoothing instead of cross-entropy (ablation) | PRG-ICFM-S · CIFAR-10 | accuracy (%): 96.18 -> 97.59 | Improved |
| use label smoothing instead of cross-entropy (ablation) | PRG-ICFM-S · Tiny-ImageNet | accuracy (%): 70.15 -> 71.12 | Improved |
| scale U-Net from 32M to 122M parameters (ablation) | PRG-ICFM · CIFAR-10 | accuracy (%): 97.35 -> 97.88 | Improved |
| scale U-Net from 32M to 122M parameters (ablation) | PRG-ICFM · Tiny-ImageNet | accuracy (%): 70.94 -> 71.80 | Improved |
| raise beta (flow-matching weight) from 1 to 100 (ablation) | PRG · CIFAR-10 / Tiny-ImageNet | accuracy: 0.973/0.712 -> 0.965/0.673 | Hurt |
| switch ODE solver from Euler to Dopri5 (ablation) | PRG-OTCFM · CIFAR-10 | accuracy (%): 97.65 -> 97.72 | No clear effect |
| use OTCFM path instead of GVP path (ablation) | PRG · CIFAR-10 | accuracy (%): 97.35 -> 97.65 | Improved |
Closest papers by meaning
- Restyling Unsupervised Concept Based Interpretable Networks with Generative Models (ICLR 2025 · similarity 0.87)
- Beyond Random Augmentations: Pretraining with Hard Views (ICLR 2025 · similarity 0.86)
- Return of Unconditional Generation: A Self-supervised Representation Generation Method (NeurIPS 2024 · similarity 0.86)
- Direct Discriminative Optimization: Your Likelihood-Based Visual Generative Model is Secretly a GAN Discriminator (ICML 2025 · similarity 0.86)
- Personalized Representation from Personalized Generation (ICLR 2025 · similarity 0.86)
- RG-GAN: Dynamic Regenerative Pruning for Data-Efficient Generative Adversarial Networks (AAAI 2024 · 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.
