EGC achieves competitive generation results compared with state-of-the-art approaches on ImageNet-1k, CelebA-HQ and LSUN Church, while achieving superior classification accuracy and robustness against adversarial attacks on CIFAR-10.

2023年4月4日 · EGC achieves competitive generation results compared with state-of-the-art approaches on ImageNet-1k, CelebA-HQ and LSUN Church, while achieving superior classification accuracy and robustness against adversarial attacks on CIFAR-10.

Motivated by the flexibility of JEM and the stability of diffusion model, we propose a novel energy-based classi-fier and generator, namely EGC, which achieves superior performance in both image classification and generation tasks using a single neural network. EGC is a classifier in the forward pass and an image generator in the back-ward pass.

EGC: Image Generation and Classification via a Single Energy-Based Model (ICCV 2023)

2023年10月5日 · EGC achieves competitive generation results compared with state-of-the-art approaches on ImageNet-1k, CelebA-HQ and LSUN Church, while achieving superior classification accuracy and robustness...

2023年4月13日 · 本文提出了一种新的能量基模型（egc），用于同时进行图像生成和分类。该模型通过对每个图像分配特定的能量值，从而能够将生成和分类两个任务统一起来。与以往的方法相比，egc模型具有更好的泛化能力，并且可以产生更逼真的生成图片。

EGC achieves competitive generation results compared with state-of-the-art approaches on ImageNet-1k, CelebA-HQ and LSUN Church, while achieving superior classification accuracy and robustness against adversarial attacks on CIFAR-10. This work marks the inaugural success in mastering both domains using a unified network parameter set.

2023年4月4日 · EGC achieves competitive generation results compared with state-of-the-art approaches on ImageNet-1k, CelebA-HQ and LSUN Church, while achieving superior classification accuracy and robustness against adversarial attacks on CIFAR-10. This work marks the inaugural success in mastering both domains using a unified network parameter set.

EGC achieves competitive generation results compared with state-of-the-art approaches on ImageNet-1k, CelebA-HQ and LSUN Church, while achieving superior classification accuracy and robustness against adversarial attacks on CIFAR-10.

EGC achieves competitive generation results compared with state-of-the-art approaches on ImageNet-1k, CelebA-HQ and LSUN Church, while achieving superior classification accuracy and robustness against adversarial attacks on CIFAR-10.