TransGHOST: Real-Time, Synthesis-Free Reconstruction of Instrument-Occluded Regions in Ophthalmic Microsurgery
TransGHOST: Real-Time, Synthesis-Free Reconstruction of Instrument-Occluded Regions in Ophthalmic Microsurgery
This research introduces TransGHOST, a non-generative, real-time reconstruction system for instrument-occluded scenes in ophthalmic microsurgery. By resampling recorded frames using geometric transforms fitted to visible regions, the method provides per-pixel confidence and provenance while abstaining on unsupported areas. Evaluated on a new veterinary clinical corpus of canine cataract surgery with pixel-level labels, TransGHOST achieves 31.89 dB masked PSNR and 0.915 masked SSIM in safe mode at median latency of 51.7 ms, outperforming LaMa by 3.45 dB on matched-mask validation and demonstrating robustness to occluded candidate references.
The study presents a causal reference-guided reconstruction pipeline for ophthalmic microsurgery that avoids generative synthesis to preserve visual provenance, achieving real-time performance on a veterinary canine cataract surgery corpus with verified occluded-region metrics.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
median latency
51.7
View original evidence
provenance and marks unsupported pixels as abstentions. We evaluate TransGHOST on a new veterinary clinical corpus of canine cataract surgery with pixel-level instrument and anatomy labels. Only the occluded region is scored, and each scoring target is verified to differ from the reference supplied to the system. Safe mode reaches 31.89 dB masked PSNR and 0.915 masked SSIM at 51.7 ms median and 85.5 ms 95th-percenti
Open source S4percentile latency (95th)
85.5
View original evidence
provenance and marks unsupported pixels as abstentions. We evaluate TransGHOST on a new veterinary clinical corpus of canine cataract surgery with pixel-level instrument and anatomy labels. Only the occluded region is scored, and each scoring target is verified to differ from the reference supplied to the system. Safe mode reaches 31.89 dB masked PSNR and 0.915 masked SSIM at 51.7 ms median and 85.5 ms 95th-percenti
Open source S4masked PSNR (veterinary corpus)
31.89
View original evidence
provenance and marks unsupported pixels as abstentions. We evaluate TransGHOST on a new veterinary clinical corpus of canine cataract surgery with pixel-level instrument and anatomy labels. Only the occluded region is scored, and each scoring target is verified to differ from the reference supplied to the system. Safe mode reaches 31.89 dB masked PSNR and 0.915 masked SSIM at 51.7 ms median and 85.5 ms 95th-percenti
Open source S4masked SSIM (veterinary corpus)
0.915
View original evidence
provenance and marks unsupported pixels as abstentions. We evaluate TransGHOST on a new veterinary clinical corpus of canine cataract surgery with pixel-level instrument and anatomy labels. Only the occluded region is scored, and each scoring target is verified to differ from the reference supplied to the system. Safe mode reaches 31.89 dB masked PSNR and 0.915 masked SSIM at 51.7 ms median and 85.5 ms 95th-percenti
Open source S4masked PSNR (held-out cohort)
29.83
View original evidence
le latency. On a pre-declared held-out cohort using instrument cutouts excluded from segmentation training, it reaches 29.83 dB masked PSNR, while matched-mask evaluation on validation shows a 3.45 dB advantage over LaMa. Region-aware reference selection improves masked PSNR by 3.60 dB over recency when candidate references can themselves be occluded. Reference-age analysis further defines the operating interval for
Open source S5masked PSNR advantage over LaMa
3.45
View original evidence
le latency. On a pre-declared held-out cohort using instrument cutouts excluded from segmentation training, it reaches 29.83 dB masked PSNR, while matched-mask evaluation on validation shows a 3.45 dB advantage over LaMa. Region-aware reference selection improves masked PSNR by 3.60 dB over recency when candidate references can themselves be occluded. Reference-age analysis further defines the operating interval for
Open source S5masked PSNR improvement over recency
3.6
View original evidence
le latency. On a pre-declared held-out cohort using instrument cutouts excluded from segmentation training, it reaches 29.83 dB masked PSNR, while matched-mask evaluation on validation shows a 3.45 dB advantage over LaMa. Region-aware reference selection improves masked PSNR by 3.60 dB over recency when candidate references can themselves be occluded. Reference-age analysis further defines the operating interval for
Open source S5
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
provenance and marks unsupported pixels as abstentions. We evaluate TransGHOST on a new veterinary clinical corpus of canine cataract surgery with pixel-level instrument and anatomy labels. Only the occluded region is scored, and each scoring target is verified to differ from the reference supplied to the system. Safe mode reaches 31.89 dB masked PSNR and 0.915 masked SSIM at 51.7 ms median and 85.5 ms 95th-percenti
Open source S4
le latency. On a pre-declared held-out cohort using instrument cutouts excluded from segmentation training, it reaches 29.83 dB masked PSNR, while matched-mask evaluation on validation shows a 3.45 dB advantage over LaMa. Region-aware reference selection improves masked PSNR by 3.60 dB over recency when candidate references can themselves be occluded. Reference-age analysis further defines the operating interval for
Open source S5
Source：arXiv Robotics — research abstracts · arxiv.org