Research map / Image generation / Image restoration
See Through Gradients: Image Batch Recovery via GradInversion
Check 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 fidelity regularization to gradient matching (ablation) | GradInversion · ImageNet1K, batch size 8 | PSNR: 10.753 -> 12.058 | Helped |
| Add pixel-wise-mean group regularization (ablation) | GradInversion with fidelity regularization · ImageNet1K, batch size 8 | PSNR: 12.058 -> 12.261 | Helped |
| Add image registration to the group consensus (ablation) | GradInversion with lazy group regularization · ImageNet1K, batch size 8 | PSNR: 12.261 -> 12.929 | Helped |
| Use MOCO V2 pretrained ResNet-50 features instead of standard ResNet-50 features (ablation) | GradInversion · ImageNet1K, batch size 8 | PSNR: 11.771 -> 12.929 | Helped |
| Use standard ResNet-18 instead of MOCO V2 ResNet-50 features (ablation) | GradInversion · ImageNet1K, batch size 8 | PSNR: 12.929 -> 10.729 | Hurt |
| Use ℓ₂ distance rather than cosine distance for gradient matching (ablation) | GradInversion · ImageNet1K gradient matching | ℓ₂ gradient distance: 5.965 -> 3.835 | Helped |
| Use column-wise minima rather than summation to identify batch labels (ablation) | Batch label restoration · ImageNet1K training set, batch size 32 | label restoration accuracy: 89.88 -> 99.29 | Helped |
| Use target-batch BN statistics rather than dataset-level BN statistics (ablation) | GradInversion · ImageNet1K, batch size 8 | PSNR: 11.235 -> 12.929 | Helped |
| Increase the batch size | GradInversion · ImageNet1K | image-specific reconstruction | Hurt |
| Invert gradients averaged over 48 images | GradInversion · ImageNet1K validation set, batch size 48 | Image Identifiability Precision: ∼ 28% | Helped |
| Select images with the largest gradient ℓ₂ norm within each class | GradInversion · ImageNet1K validation set, batch size 48 | Image Identifiability Precision: nearly doubled | Helped |
Closest papers by meaning
- SPEAR: Exact Gradient Inversion of Batches in Federated Learning (NeurIPS 2024 · similarity 0.88)
- High-Fidelity Gradient Inversion in Distributed Learning (AAAI 2024 · similarity 0.87)
- DAGER: Exact Gradient Inversion for Large Language Models (NeurIPS 2024 · similarity 0.87)
- GRAIN: Exact Graph Reconstruction from Gradients (ICLR 2025 · similarity 0.86)
- HyperNet Fields: Efficiently Training Hypernetworks without Ground Truth by Learning Weight Trajectories (CVPR 2025 · similarity 0.85)
- Preserving Deep Representations in One-Shot Pruning: A Hessian-Free Second-Order Optimization Framework (ICLR 2025 · 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.