Research map / Continual learning / Online and replay-based
Lifelong GAN: Continual Learning for Conditional Image Generation
Mengyao Zhai, Lei Chen, Frederick Tung, Jiawei He, Megha Nawhal, Greg Mori
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 Lifelong GAN distillation instead of sequential fine-tuning, MNIST segmentation | BicycleGAN · MNIST segmentation (image-conditioned) | Acc: 58.02 -> 95.90 | Improved |
| add Lifelong GAN distillation instead of sequential fine-tuning, image-to-image | BicycleGAN · Edges->shoes then segmentations->facades | Acc: 39.72 -> 98.93 | Improved |
| add Lifelong GAN distillation instead of sequential fine-tuning, MNIST label-conditioned | BicycleGAN · MNIST digits (label-conditioned) | Acc: 21.59 -> 97.52 | Improved |
| use distillation instead of memory replay, flower generation | BicycleGAN · Flower (label-conditioned) | Acc: 87.6 -> 98.4 | Improved |
| use distillation instead of memory replay, flower generation | BicycleGAN · Flower (label-conditioned) | r-Acc: 60.4 -> 85.6 | Improved |
| use distillation instead of memory replay, MNIST digits | BicycleGAN · MNIST digits (label-conditioned) | Acc: 97.54 -> 97.52 | No clear effect |
| use distillation instead of memory replay, MNIST digits | BicycleGAN · MNIST digits (label-conditioned) | r-Acc: 85.57 -> 87.77 | Improved |
| add montage and swap auxiliary data (ablation) | Lifelong GAN · Edges->shoes then segmentations->facades | Acc: 66.78 -> 98.93 | Improved |
| add montage and swap auxiliary data (ablation) | Lifelong GAN · Edges->shoes then segmentations->facades | r-Acc: 59.76 -> 99.37 | Improved |
| add swap operation (ablation) | Lifelong GAN · Edges->shoes then segmentations->facades | r-Acc: 86.80 -> 99.37 | Improved |
| add swap operation (ablation) | Lifelong GAN · Edges->shoes then segmentations->facades | Acc: 97.62 -> 98.93 | Improved |
| add montage and swap, LPIPS closer to real (ablation) | Lifelong GAN · Edges->shoes then segmentations->facades | LPIPS: 0.518 -> 0.417 | Improved |
Closest papers by meaning
- Hyper-LifelongGAN: Scalable Lifelong Learning for Image Conditioned Generation (CVPR 2021 · similarity 0.94)
- GAN Memory with No Forgetting (NeurIPS 2020 · similarity 0.92)
- Piggyback GAN: Efficient Lifelong Learning for Image Conditioned Generation (ECCV 2020 · similarity 0.91)
- Memory Replay GANs: Learning to Generate New Categories without Forgetting (NeurIPS 2018 · similarity 0.91)
- Recall-Oriented Continual Learning with Generative Adversarial Meta-Model (AAAI 2024 · similarity 0.89)
- Task-Free Continual Generation and Representation Learning via Dynamic Expansionable Memory Cluster (AAAI 2024 · 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.
