# Harmonizing Maximum Likelihood with GANs for Multimodal Conditional Generation

arXiv: https://arxiv.org/abs/1902.09225 · HTML: https://mlsanity.com/paper/1902.09225/

## Key facts

- Cited by 7 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 25 Feb 2019.

## Built on by (most cited first, 7 of 7)

- [Collapse by Conditioning: Training Class-conditional GANs with Limited Data](https://mlsanity.com/paper/2201.06578/) (2022)
- [CoPE: Conditional image generation using Polynomial Expansions](https://mlsanity.com/paper/2104.05077/) (2021)
- [Multi-modal Latent Diffusion](https://mlsanity.com/paper/2306.04445/) (2023)
- [Stay Positive: Non-Negative Image Synthesis for Augmented Reality](https://mlsanity.com/paper/2202.00659/) (2022)
- [Auto-Encoded Supervision for Perceptual Image Super-Resolution](https://mlsanity.com/paper/2412.00124/) (2024)
- [Channel-wise Noise Scheduled Diffusion for Inverse Rendering in Indoor Scenes](https://mlsanity.com/paper/2503.09993/) (2025)
- [MaCo-GAN: Manifold-Contrastive Adversarial Learning for Single Image Super-Resolution](https://mlsanity.com/paper/2606.05068/) (2026)

## Abstract (arXiv)

Recent advances in conditional image generation tasks, such as image-to-image translation and image inpainting, are largely accounted to the success of conditional GAN models, which are often optimized by the joint use of the GAN loss with the reconstruction loss. However, we reveal that this training recipe shared by almost all existing methods causes one critical side effect: lack of diversity in output samples. In order to accomplish both training stability and multimodal output generation, we propose novel training schemes with a new set of losses named moment reconstruction losses that simply replace the reconstruction loss. We show that our approach is applicable to any conditional generation tasks by performing thorough experiments on image-to-image translation, super-resolution and image inpainting using Cityscapes and CelebA dataset. Quantitative evaluations also confirm that our methods achieve a great diversity in outputs while retaining or even improving the visual fidelity of generated samples.