Research map / Image generation / Autoregressive and latent generation
LIFT: Latent Implicit Functions for Task- and Data-Agnostic Encoding
Amirhossein Kazerouni, Soroush Mehraban, Michael Brudno, Babak Taati
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 intermediate latent to global+local (ablation) | LIFT · CelebA-HQ reconstruction | Test PSNR: 39.62 -> 40.91 | Improved |
| add local latent to global-only (ablation) | LIFT · CelebA-HQ reconstruction | Test PSNR: 24.82 -> 39.62 | Improved |
| add hierarchical latent generator (HLG) (ablation) | LIFT · CelebA-HQ reconstruction | Test PSNR: 39.28 -> 40.91 | Improved |
| add hierarchical latent generator (HLG) (ablation) | LIFT · CelebA-HQ reconstruction | rFID: 2.70 -> 2.22 | Improved |
| increase Tinner from 1 to 3 (ablation) | LIFT · CelebA-HQ reconstruction | Test PSNR: 33.25 -> 40.17 | Improved |
| increase Tinner from 3 to 4 (ablation) | LIFT · CelebA-HQ reconstruction | Test PSNR: 40.17 -> 40.28 | No clear effect |
| double local latent spatial size from 4x4 to 8x8 (ablation) | LIFT · CelebA-HQ reconstruction | Test PSNR: 30.27 -> 40.42 | Improved |
| enlarge intermediate latent from 2x2 to 4x4 (ablation) | LIFT · CelebA-HQ reconstruction | Test PSNR: 40.42 -> 40.91 | Improved |
| double channels of all latents (ablation) | LIFT · CelebA-HQ reconstruction | Test PSNR: 40.91 -> 49.86 | Improved |
| increase P-MLP depth from 1x256 to 16x16 (narrower, deeper) (ablation) | LIFT P-MLP · CelebA-HQ reconstruction | Test PSNR: 40.25 -> 35.16 | Hurt |
| increase augmentations from 1 to 10 (ablation) | LIFT · CIFAR-10 classification | Top-1 accuracy: 91.04 -> 95.47 | Improved |
| use LIFT multiscale latent instead of global Functa latent | LIFT · CIFAR-10 classification | Top-1 accuracy: 68.30 -> 95.47 | Improved |
Closest papers by meaning
- Hyper-Transforming Latent Diffusion Models (ICML 2025 · similarity 0.86)
- Few-shot Implicit Function Generation via Equivariance (CVPR 2025 · similarity 0.86)
- MIRE: Matched Implicit Neural Representations (CVPR 2025 · similarity 0.86)
- VI^3NR: Variance Informed Initialization for Implicit Neural Representations (CVPR 2025 · similarity 0.85)
- RECAST: Reparameterized, Compact weight Adaptation for Sequential Tasks (ICLR 2025 · similarity 0.85)
- Beyond Linear Activation Steering: Invertible Latent Transformations for Controlling LLM Behavior (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.
