RefRoute: Decoupling Conditioning Cost from References via Compact Residual Conditioning and Spatial Routing

Key points

• Multi-reference image generation requires preserving the appearance of multiple subjects while composing them into a coherent scene.1
• Compact residual conditioning combines low-resolution latent tokens with lightweight residual features extracted from full-resolution pixels, reducing reference token counts while retaining fine-grained appearance cues.1
• We further introduce RefRoute-Data for training many-reference generation models and ManyRef100, a benchmark spanning human, object, and mixed compositions with 10-17 references.1
• These results establish compact reference representations and spatially routed attention as an effective approach to scalable many-reference image generation.1

Sources (1)

• [1]RefRoute: Decoupling Conditioning Cost from References via Compact Residual Conditioning and Spatial Routing
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 6, 04:16 AM

  We present RefRoute, a framework that addresses both reference representation cost and attention overhead through two complementary mechanisms.

  Multi-reference image generation requires preserving the appearance of multiple subjects while composing them into a coherent scene.

Extractive summary: sentences quoted from the sources.