Search papers, labs, and topics across Lattice.
The paper introduces GPEReg-Net, a scene-appearance disentanglement framework for registering forward and backward scan lines in bidirectional optical-resolution photoacoustic microscopy (OR-PAM). GPEReg-Net disentangles domain-invariant scene features from domain-specific appearance codes using Adaptive Instance Normalization (AdaIN) and incorporates a Global Position Encoding (GPE) module to exploit temporal structure via cross-frame attention. Experiments on the OR-PAM-Reg-4K benchmark demonstrate that GPEReg-Net achieves state-of-the-art registration performance, improving SSIM by 3.8% and PSNR by 1.99dB compared to existing methods.
By disentangling scene features from domain-specific appearances and leveraging temporal context, GPEReg-Net achieves state-of-the-art registration performance in bidirectional photoacoustic microscopy, surpassing existing methods by a significant margin.
High-speed optical-resolution photoacoustic microscopy (OR-PAM) with bidirectional raster scanning doubles imaging speed but introduces coupled domain shift and geometric misalignment between forward and backward scan lines. Existing registration methods, constrained by brightness constancy assumptions, achieve limited alignment quality, while recent generative approaches address domain shift through complex architectures that lack temporal awareness across frames. We propose GPEReg-Net, a scene-appearance disentanglement framework that separates domain-invariant scene features from domain-specific appearance codes via Adaptive Instance Normalization (AdaIN), enabling direct image-to-image registration without explicit deformation field estimation. To exploit temporal structure in sequential acquisitions, we introduce a Global Position Encoding (GPE) module that combines learnable position embeddings with sinusoidal encoding and cross-frame attention, allowing the network to leverage context from neighboring frames for improved temporal coherence. On the OR-PAM-Reg-4K benchmark (432 test samples), GPEReg-Net achieves NCC of 0.953, SSIM of 0.932, and PSNR of 34.49dB, surpassing the state-of-the-art by 3.8% in SSIM and 1.99dB in PSNR while maintaining competitive NCC. Code is available at https://github.com/JiahaoQin/GPEReg-Net.
