# CSG0: Continual Urban Scene Generation with Zero Forgetting

arXiv: https://arxiv.org/abs/2112.03252 · HTML: https://mlsanity.com/paper/2112.03252/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 23 of its 27 references that are in the map.
- First posted on arXiv 6 Dec 2021.

## Builds on (most cited first, 23 of 23)

- [Overcoming catastrophic forgetting in neural networks](https://mlsanity.com/paper/1612.00796/) (2016)
- [Image-to-Image Translation with Conditional Adversarial Networks](https://mlsanity.com/paper/1611.07004/) (2016)
- [The Cityscapes Dataset for Semantic Urban Scene Understanding](https://mlsanity.com/paper/1604.01685/) (2016)
- [Pyramid Scene Parsing Network](https://mlsanity.com/paper/1612.01105/) (2016)
- [Arbitrary Style Transfer in Real-time with Adaptive Instance Normalization](https://mlsanity.com/paper/1703.06868/) (2017)
- [Semantic Image Synthesis with Spatially-Adaptive Normalization](https://mlsanity.com/paper/1903.07291/) (2019)
- [Instance Normalization: The Missing Ingredient for Fast Stylization](https://mlsanity.com/paper/1607.08022/) (2016)
- [High-Resolution Image Synthesis and Semantic Manipulation with Conditional GANs](https://mlsanity.com/paper/1711.11585/) (2017)
- [Playing for Data: Ground Truth from Computer Games](https://mlsanity.com/paper/1608.02192/) (2016)
- [A Learned Representation For Artistic Style](https://mlsanity.com/paper/1610.07629/) (2016)
- [IDD: A Dataset for Exploring Problems of Autonomous Navigation in Unconstrained Environments](https://mlsanity.com/paper/1811.10200/) (2018)
- [Transferring GANs: generating images from limited data](https://mlsanity.com/paper/1805.01677/) (2018)
- [Freeze the Discriminator: a Simple Baseline for Fine-Tuning GANs](https://mlsanity.com/paper/2002.10964/) (2020)
- [You Only Need Adversarial Supervision for Semantic Image Synthesis](https://mlsanity.com/paper/2012.04781/) (2020)
- [Learning to Predict Layout-to-image Conditional Convolutions for Semantic Image Synthesis](https://mlsanity.com/paper/1910.06809/) (2019)
- [MineGAN: effective knowledge transfer from GANs to target domains with few images](https://mlsanity.com/paper/1912.05270/) (2019)
- [Image Generation From Small Datasets via Batch Statistics Adaptation](https://mlsanity.com/paper/1904.01774/) (2019)
- [GAN Memory with No Forgetting](https://mlsanity.com/paper/2006.07543/) (2020)
- [Lifelong GAN: Continual Learning for Conditional Image Generation](https://mlsanity.com/paper/1907.10107/) (2019)
- [Memory Replay GANs: learning to generate images from new categories without forgetting](https://mlsanity.com/paper/1809.02058/) (2018)
- [Continual Learning in Generative Adversarial Nets](https://mlsanity.com/paper/1705.08395/) (2017)
- [Piggyback GAN: Efficient Lifelong Learning for Image Conditioned Generation](https://mlsanity.com/paper/2104.11939/) (2021)
- [Efficient Conditional GAN Transfer with Knowledge Propagation across Classes](https://mlsanity.com/paper/2102.06696/) (2021)

## Abstract (arXiv)

With the rapid advances in generative adversarial networks (GANs), the visual quality of synthesised scenes keeps improving, including for complex urban scenes with applications to automated driving. We address in this work a continual scene generation setup in which GANs are trained on a stream of distinct domains; ideally, the learned models should eventually be able to generate new scenes in all seen domains. This setup reflects the real-life scenario where data are continuously acquired in different places at different times. In such a continual setup, we aim for learning with zero forgetting, IE, with no degradation in synthesis quality over earlier domains due to catastrophic forgetting. To this end, we introduce a novel framework that not only (i) enables seamless knowledge transfer in continual training but also (ii) guarantees zero forgetting with a small overhead cost. While being more memory efficient, thanks to continual learning, our model obtains better synthesis quality as compared against the brute-force solution that trains one full model for each domain. Especially, under extreme low-data regimes, our approach outperforms the brute-force one by a large margin.