AdvFD: Boosting Visual Generation via Adversarial Fr'echet Distance Loss

arXiv:2608.11205v1 Announce Type: new Abstract: Frechet distance has recently emerged as an effective distribution-level objective for generator post-training, complementing the conventional sample-level diffusion and flow-matching losses. However, directly optimizing Frechet objectives can cause Frechet hacking. The target metrics keep improving, but visual quality and Frechet alignment in other feature spaces may stagnate or deteriorate. We attribute this failure to the static pretrained feature spaces used by existing Frechet losses. These feature spaces provide incomplete and fixed views of the differences between real and generated distributions. To address this limitation, we propose Adversarial Frechet Distance (AdvFD), which complements the static representation targets in FD-Loss with a calibrated adversarially learned representation. AdvFD augments the original static Frechet objective with a learnable representation that adversarially maximizes the Frechet discrepancy between real and generated samples, while the generator minimizes the same discrepancy in the resulting adaptive feature space. To prevent the adversarial representation from trivially increasing the objective through feature amplification, we further introduce real-feature whitening, which normalizes its scale and covariance geometry and stabilizes the min--max optimization. Extensive experiments show that AdvFD consistently improves one-step generator post-training across both JiT and pMF backbones and across different model scales.

Related

• FAIL: Flow Matching Adversarial Imitation Learning for Image Generation
• PISCES: Annotation-free Text-to-Video Post-Training via Optimal Transport-Aligned Rewards
• Sample-Adaptive Latent Rewards for Uncertainty-Guided Diffusion Post-Training
• Pave-GRPO: Beyond Instantaneous Guidance through Principled Average Velocity Decomposition

Source: arXiv cs.CV | 2026-08-12