Research map / Image generation / GANs
Meta Internal 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 |
|---|---|---|---|
| Train the generator and discriminator weights through image-conditioned hypernetworks instead of fitting a separate model per image | Meta Internal Learning · Places-50 / LSUN-50 | Training time per image (minutes): 60 | Helped |
| Train with the dataset rather than single-image training | Meta Internal Learning · Places-50 / LSUN-50 | SIFID: 0.03/0.11 -> 0.05/0.11 | Mixed |
| Train with the dataset rather than single-image training | Meta Internal Learning · Places-50 / LSUN-50 | mSIFID: 0.06/0.19 -> 0.07/0.16 | Mixed |
| Increase training data from Valley500 to Valley5000 | Meta Internal Learning · Valley test set | SIFID: 0.47 -> 0.41 | Helped |
| Increase training data from Valley500 to Valley5000 | Meta Internal Learning · Valley test set | mSIFID: 2.47 -> 1.52 | Helped |
| Increase mini-batch size from 1 to 5 | Meta Internal Learning · Single mini-batch training | SIFID: 0.03 -> 0.04 | Mixed |
| Freeze the ResNet34 encoder and use a 1-layer projection rather than end-to-end training (ablation) | Meta Internal Learning · Places-50 | SIFID: 0.05 -> 0.26 | Hurt |
| Freeze the ResNet34 encoder and use a 3-layer projection rather than end-to-end training (ablation) | Meta Internal Learning · Places-50 | SIFID: 0.05 -> 0.14 | Hurt |
| Freeze the ResNet34 encoder and use a 5-layer projection rather than end-to-end training (ablation) | Meta Internal Learning · Places-50 | SIFID: 0.05 -> 0.17 | Hurt |
| Use a shared discriminator instead of an image-conditioned hyper-discriminator (ablation) | Meta Internal Learning · Multi-image training | Image-conditioned patch generation | Hurt |
| Share the generator and discriminator feature extractor (ablation) | Meta Internal Learning · Adversarial training | Discriminator effectiveness | Hurt |
Closest papers by meaning
- PetsGAN: Rethinking Priors for Single Image Generation (AAAI 2022 · similarity 0.88)
- Patchwise Generative ConvNet: Training Energy-Based Models From a Single Natural Image for Internal Learning (CVPR 2021 · similarity 0.87)
- Instance-Conditioned GAN (NeurIPS 2021 · similarity 0.87)
- Meta Convolutional Neural Networks for Single Domain Generalization (CVPR 2022 · similarity 0.86)
- Non-Adversarial Image Synthesis With Generative Latent Nearest Neighbors (CVPR 2019 · similarity 0.86)
- On Leveraging Pretrained GANs for Generation with Limited Data (ICML 2020 · similarity 0.86)
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.