Research map / Image generation / Autoregressive and latent generation
Generative Models as a Data Source for Multiview Representation Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 Gaussian latent views to generated samples alongside SimCLR pixel augmentations (ablation) | BigBiGAN representations · ImageNet1000 | Top-1 accuracy: 35.69 -> 42.58 | Helped |
| Add Gaussian latent views to generated samples alongside SimCLR pixel augmentations (ablation) | BigBiGAN representations · VOC07 classification | AP: 0.57 -> 0.64 | Helped |
| Replace real training images with StyleGAN2 samples and add Gaussian latent views | SimCLR representations · Stanford Car classification | Top-1 accuracy: 39.68 -> 49.79 | Helped |
| Add Gaussian latent views to StyleGAN2-generated images while retaining pixel augmentations (ablation) | StyleGAN2 LSUN Car representations · ImageNet1000 | Top-1 accuracy: 28.33 -> 30.06 | Helped |
| Use Gaussian latent views instead of steered latent views, with pixel augmentations (ablation) | BigBiGAN representations · ImageNet1000 | Top-1 accuracy: 41.78 -> 42.58 | Helped |
| Add Gaussian latent views without pixel-space augmentations (ablation) | BigBiGAN representations · ImageNet1000 | Top-1 accuracy: 35.69 -> 28.88 | Hurt |
| Replace Gaussian latent views with steered latent views, without pixel augmentations (ablation) | BigBiGAN representations · ImageNet1000 | Top-1 accuracy: 28.88 -> 26.52 | Hurt |
| Use Gaussian latent views rather than steering with pixel augmentations (ablation) | BigGAN supervised contrastive representations · ImageNet1000 | Top-1 accuracy: 49.25 -> 46.43 | Hurt |
| Add Gaussian latent views to class-conditional samples with pixel augmentations (ablation) | BigGAN supervised contrastive representations · ImageNet1000 | Top-1 accuracy: 48.19 -> 46.43 | Hurt |
| Add steered latent views to class-conditional samples with pixel augmentations (ablation) | BigGAN supervised contrastive representations · ImageNet1000 | Top-1 accuracy: 48.19 -> 49.25 | Helped |
| Use independent latent samples as positive views, retaining pixel augmentations (ablation) | BigGAN supervised contrastive representations · ImageNet1000 | Top-1 accuracy: 48.19 -> 48.97 | Helped |
| Replace supervised contrastive learning with latent-code inversion | BigGAN representations · ImageNet1000 | Top-1 accuracy: 48.19 -> 31.56 | Hurt |
Closest papers by meaning
- Can Generative Models Improve Self-Supervised Representation Learning? (AAAI 2025 · similarity 0.90)
- Personalized Representation from Personalized Generation (ICLR 2025 · similarity 0.89)
- Restyling Unsupervised Concept Based Interpretable Networks with Generative Models (ICLR 2025 · similarity 0.88)
- Compositional Scene Understanding through Inverse Generative Modeling (ICML 2025 · similarity 0.88)
- Learning Visual Generative Priors without Text (CVPR 2025 · similarity 0.88)
- Boosting Generative Image Modeling via Joint Image-Feature Synthesis (NeurIPS 2025 · similarity 0.88)
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.