Search papers, labs, and topics across Lattice.
The paper introduces PostureObjectStitch, a novel image synthesis approach for generating anomaly images in industrial assembly scenarios by explicitly modeling component pose and assembly relationships. They decouple multi-view images into high-frequency, texture, and RGB features, and use a feature temporal modulation mechanism within a diffusion model to generate images progressively. The method incorporates a conditional loss to emphasize critical industrial elements and a geometric prior to enforce correct component positioning, leading to improved performance on multiple datasets.
Synthesizing realistic anomaly images for industrial assembly is now possible thanks to a diffusion model that respects component pose and assembly relationships.
Image generation technology can synthesize condition-specific images to supplement real-world industrial anomaly data and enhance anomaly detection model performance. Existing generation techniques rarely account for the pose and orientation of industrial components in assembly, making the generated images difficult to utilize for downstream application. To solve this, we propose a novel image synthesis approach, called PostureObjectStitch, that achieves accurate generation to meet the requirement of industrial assembly. A condition decoupling approach is introduced to separate input multi-view images into high-frequency, texture, and RGB features. The feature temporal modulation mechanism adapts these features across diffusion model time-steps, enabling progressive generation from coarse to fine details while maintaining consistency. To ensure semantic accuracy, we introduce a conditional loss that enhances critical industrial elements and a geometric prior that guides component positioning for correct assembly relationships. Comprehensive experimental results on the MureCom dataset, our newly contributed DreamAssembly dataset, and the downstream application validate the outstanding performance of our method.
Tsinghua AI