Trajectory Consistency Distillation: Improved Latent Consistency Distillation by Semi-Linear Consistency Function with Trajectory Mapping

2024/02/29 by Jianbin Zheng, Zheng Jianbin, Minghui Hu +13 · 1 voice · 9 citations
Biochemistry, Genetics and Molecular Biology · Computer Science · #Advanced Image and Video Retrieval Techniques #Cell Image Analysis Techniques #Domain Adaptation and Few-Shot Learning #cs.CV

paper · pdf · doi:10.48550/arxiv.2402.19159

openalex publication_date 2024/02/29 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

Latent Consistency Model (LCM) extends the Consistency Model to the latent space and leverages the guided consistency distillation technique to achieve impressive performance in accelerating text-to-image synthesis. However, we observed that LCM struggles to generate images with both clarity and detailed intricacy. Consequently, we introduce Trajectory Consistency Distillation (TCD), which encompasses trajectory consistency function and strategic stochastic sampling. The trajectory consistency function diminishes the parameterisation and distillation errors by broadening the scope of the self-consistency boundary condition with trajectory mapping and endowing the TCD with the ability to accurately trace the entire trajectory of the Probability Flow ODE in semi-linear form with an Exponential Integrator. Additionally, strategic stochastic sampling provides explicit control of stochastic and circumvents the accumulated errors inherent in multi-step consistency sampling. Experiments demonstrate that TCD not only significantly enhances image quality at low NFEs but also yields more detailed results compared to the teacher model at high NFEs.

Cited by

• UniLumos: Fast and Unified Image and Video Relighting with Physics-Plausible Feedback
• Self-Evaluation Unlocks Any-Step Text-to-Image Generation
• Flash-DMD: Towards High-Fidelity Few-Step Image Generation with Efficient Distillation and Joint Reinforcement Learning
• Masked Auto-Regressive Variational Acceleration: Fast Inference Makes Practical Reinforcement Learning
• Hybrid Consistency Policy: Decoupling Multi-Modal Diversity and Real-Time Efficiency in Robotic Manipulation
• EchoDistill: Bidirectional Concept Distillation for One-Step Diffusion Personalization
• A multi-temporal multi-spectral attention-augmented deep convolution neural network with contrastive learning for crop yield prediction
• Q-Sched: Pushing the Boundaries of Few-Step Diffusion Models with Quantization-Aware Scheduling
• LawDIS: Language-Window-based Controllable Dichotomous Image Segmentation

Discussions

• Trajectory Consistency Distillation [lemmy, 21 points, 0 comments]

Related

• SegmentDreamer: Towards High-fidelity Text-to-3D Synthesis with Segmented Consistency Trajectory Distillation
• Hyper-SD: Trajectory Segmented Consistency Model for Efficient Image Synthesis
• Reward Guided Latent Consistency Distillation
• Image-Free Timestep Distillation via Continuous-Time Consistency with Trajectory-Sampled Pairs
• Target-Driven Distillation: Consistency Distillation with Target Timestep Selection and Decoupled Guidance