Research map / Learning theory and algorithms / Matrix and tensor methods
GAN-Based Projector for Faster Recovery With Convergence Guarantees in Linear Inverse Problems
Ankit Raj, Yuqi Li, Yoram Bresler
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 |
|---|---|---|---|
| replace inner-loop GAN inversion with a learned network projector (no Jacobian of G) | PGD with GAN prior · CelebA compressed sensing run time | speedup: 64x to 80x over CSGM, m=200 to 2000 | Improved |
| use G-based designed measurement matrix instead of random Gaussian | NPGD · Compressed sensing (MNIST, CelebA) | reconstruction error and SSIM: 5-10x fewer measurements for comparable recovery | Improved |
| increase latent dimension k of the projector (ablation) | NPGD projector · MNIST idempotence error | relative idempotence error: saturates around k=100 | Improved |
| add noise sigma=1 to train projector instead of low sigma (ablation) | NPGD projector · projector training | projection quality | Improved |
| increase downsampling factor f in super-resolution | NPGD · MNIST super-resolution | reconstruction quality: f=2,3,4 | Hurt |
| increase inpainting mask size | NPGD · MNIST inpainting | reconstruction quality: masks 6, 10, 14 | Hurt |
Closest papers by meaning
- Invertible generative models for inverse problems: mitigating representation error and dataset bias (ICML 2020 · similarity 0.87)
- Image-Adaptive GAN Based Reconstruction (AAAI 2020 · similarity 0.87)
- Task-Aware Compressed Sensing With Generative Adversarial Networks (AAAI 2018 · similarity 0.86)
- One Network to Solve Them All -- Solving Linear Inverse Problems Using Deep Projection Models (ICCV 2017 · similarity 0.86)
- pcaGAN: Improving Posterior-Sampling cGANs via Principal Component Regularization (NeurIPS 2024 · similarity 0.86)
- A Restoration Network as an Implicit Prior (ICLR 2024 · similarity 0.84)
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.
