Revisiting the Evaluation of Image Synthesis with GANs Mengping Y ang 1, Ceyuan Y ang
–Neural Information Processing Systems
Unlike most vision tasks that have per-sample ground-truth, image synthesis tasks target generating unseen data and hence are usually evaluated through a distributional distance between one set of real samples and another set of generated samples.
Neural Information Processing Systems
Feb-8-2026, 16:37:22 GMT