能量引導流匹配

摘要

像素空間生成模型繞過了有損的潛在壓縮，但卻需要在高維空間中同時學習全域結構與精細細節。標準的流匹配是將雜訊向固定的乾淨影像端點進行插值，使得頻譜演化必須被隱式地學習。在本論文中，我們提出了能量引導流匹配（Energy-Guided Flow Matching, EG-FM），透過移動端點來明確建模由粗到細的生成軌跡。具體而言，EG-FM 以熱核濾波端點取代固定端點，該端點會從低頻影像平滑地演化至乾淨影像。移動端點中的高頻訊號比例，由影像特定的能量引導調度釋放，從而在流匹配中重新定向速度。我們的框架無需調整骨幹網路與訓練資料，且在訓練與推論階段僅帶來可忽略的成本。在實驗中，EG-FM 在 ImageNet 類別條件影像生成任務上，以更少的時期持續取得較低的 FID，在 256×256 解析度下，200 個時期達到 FID 1.55，600 個時期達到 1.45。我們在 512×512 解析度設定下繼續訓練生成任務，僅經過 40 個高解析度適應時期，便達到 FID 1.58。此外，我們將 EG-FM 應用於文字轉影像生成，在 GenEval 得分上達到 0.85，在 DPG-Bench 上達到 83.9。程式碼已公開於 https://github.com/ysng123/EG-FM。

English

Pixel-space generative models bypass lossy latent compression, yet necessitate joint learning of global structure and fine-grained details in a high-dimensional space. Standard flow matching interpolates noise toward a fixed clean-image endpoint, leaving the spectral evolution to be learned implicitly. In this paper, we introduce Energy-Guided Flow Matching(EG-FM) that explicitly models a coarse-to-fine generative trajectory by moving endpoint. Specifically, EG-FM replaces the fixed endpoint with a heat-kernel-filtered endpoint that evolves smoothly from low-frequency image to clean image. The fraction of high-frequency signal in moving endpoint is released by an image-specific energy-guided scheduling, leading to the re-targeting of velocity in flow matching. Our framework requires no adaptation of the backbone and training data, bringing negligible cost on the training and inference stages. In our experiment, EG-FM consistently achieves lower FID on the ImageNet class-conditional image generation task at 256 times 256 with fewer epochs, reaching an FID of 1.55 at 200 epochs and 1.45 at 600 epochs. We continue training the generation task on the setting of 512 times 512 resolution, yielding a FID of 1.58 after only 40 high-resolution adaptation epochs. Furthermore, we transfer EG-FM on text-to-image generation and achieve 0.85 on GenEval score and 83.9 on DPG-Bench. Code is available at https://github.com/ysng123/EG-FM.