Research map / Image generation / GANs
Deep Compressed Sensing
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 |
|---|---|---|---|
| Use meta-learned latent optimisation with 3 gradient steps instead of the baseline's lengthy iterative reconstruction | Deep compressed sensing reconstruction · MNIST test and CelebA test | reconstruction loss: MNIST: 54.8 -> 5.3 (10 measurements), 17.2 -> 3.4 (25 measurements); CelebA: 156.8 -> 23.4 (20 measurements), 82.3 -> 18 | Helped |
| Use random linear measurements rather than neural-network measurements, with the measurement function fixed (ablation) | DCS reconstruction · MNIST and CelebA test | reconstruction loss: MNIST: 12.5 -> 10.8 (10 measurements), 10.2 -> 6.9 (25 measurements); CelebA: 46.1 -> 34.7 (20 measurements), 41.2 -> 27 | Helped |
| Learn the measurement function rather than keep measurements fixed (ablation) | DCS reconstruction · MNIST and CelebA test | reconstruction loss: MNIST: 10.8 -> 6.5 and 6.9 -> 4.; CelebA: 34.7 -> 26.2 and 27.1 -> 20.5 | Helped |
| Replace learned linear measurements with learned neural-network measurements (ablation) | DCS reconstruction · MNIST and CelebA test | reconstruction loss: MNIST: 6.5 -> 5.3 and 4. -> 3.4; CelebA: 26.2 -> 23.4 and 20.5 -> 18.5 | Helped |
| Add latent optimisation to vanilla GAN training (ablation) | CS-GAN · CIFAR10 | Inception Score | Helped |
| Add latent optimisation to vanilla GAN training (ablation) | CS-GAN · CIFAR10 | FID | Helped |
| Add latent optimisation to SN-GAN (ablation) | SN-GAN · CIFAR10 | Inception Score: 7.34 ± 0.07 -> 7.80 ± 0.05 | Helped |
| Add latent optimisation to SN-GAN (ablation) | SN-GAN · CIFAR10 | FID: 29.53 ± 0.36 -> 23.13 ± 0.50 | Helped |
| Use 3 or 5 latent optimisation steps rather than 0 (ablation) | CS-GAN · MNIST | mode collapse | Helped |
| Use 3 latent optimisation steps rather than 0 (ablation) | CS-GAN · CIFAR10 across 144 hyper-parameter settings | training stability | Helped |
| Vary the latent optimisation cost weight β from 1.0 to 10.0 (ablation) | CS-GAN · CS-GAN training | training performance | No clear effect |
Closest papers by meaning
- Task-Aware Compressed Sensing With Generative Adversarial Networks (AAAI 2018 · similarity 0.92)
- Compressed Sensing using Generative Models (ICML 2017 · similarity 0.91)
- Modeling Sparse Deviations for Compressed Sensing using Generative Models (ICML 2018 · similarity 0.90)
- On the Power of Compressed Sensing with Generative Models (ICML 2020 · similarity 0.90)
- Compressive sensing with un-trained neural networks: Gradient descent finds a smooth approximation (ICML 2020 · similarity 0.89)
- SEGAN: Structure-Enhanced Generative Adversarial Network for Compressed Sensing MRI Reconstruction (AAAI 2019 · 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.