Low-light image enhancement is crucial for accurately interpreting images captured under low-lighting conditions. Existing low-light enhancement methods based on diffusion models have demonstrated effectiveness; however, they suffer from slow training processes and less structured guidance during optimization. We propose a navel patch-wise-based diffusion model, which introduces the low curvature reverse trajectory to ensure stable parameter optimization and uncertainty guidance in the diffusion training process. Specifically, we randomly select patches of varying sizes from the entire image and apply patch-wise optimization between the generated image and the ground truth to enforce a stable optimization path in the diffusion model. Additionally, within each patch-wise region, an uncertainty network estimates the uncertainty, which is then integrated as a weighting factor in the diffusion process to balance areas of abrupt change in the image. Experimental evaluations on various datasets demonstrate that our method achieves significant improvements, particularly in experiments with real-world images. These results indicate that the proposed patch-wise-based diffusion model enhancements are effective.
Interface interaction mechanics analysis is of great significance for robot-assisted insertion surgery in minimally invasive surgery and therapy. Previous work indicates that the accurate modeling of soft tissue puncture forces plays a crucial role in surgical planning, robotic needle insertion, and biomechanical simulation, which can give insights useful for physicians to guide and operate assisted robots. The objective of this study is to develop a dynamic multi-component force model that integrates cutting force, stiffness resistance, and frictional interaction to characterize needle–soft tissue interaction during puncture. A dynamic force model is proposed, and a lateral periodic disturbance mechanism is introduced into the simulation framework in order to enhance the robustness and realism of the model under micro-manipulation scenarios. The model has been validated using a series of controlled puncture experiments on porcine liver and renal tissues under varying insertion angles (15°, 30°, 45°) and speeds (0.5 mm/s, 1.5 mm/s, 2.5 mm/s). Corresponding finite element simulations were also conducted using ANSYS software. The agreement between simulation and experiment has been quantitatively evaluated by comparing force–depth and force–time curves, and the statistical significance of the impact of angle and speed on puncture forces has been assessed using ANOVA and Tukey’s HSD tests. Quantitative comparison demonstrated strong consistency, with the optimal case reaching a coefficient of determination (R2) value of 0.96 and Root Mean Square Error (RMSE) below 0.13 N after incorporating a 0.05 mm lateral perturbation. Statistical analysis confirmed the impact of angle and speed on puncture force responses (p < 0.05). Furthermore, comparative analysis revealed that porcine liver exhibits more consistent biomechanical behavior than renal tissue, particularly under perturbation-enhanced simulation. This study successfully establishes a dynamic multi-component force model for soft tissue puncture, validated with high fidelity against experimental data. The incorporated lateral disturbance mechanism enhanced the model’s realism. This work can provide a reliable foundation for the future design of intelligent robot-assisted puncture systems and high-fidelity simulation-based training platforms.
Monocular reconstruction under severe low light is ill-posed because underexposure removes repeatable texture, view-dependent camera processing violates photometric consistency, and pose or depth errors can be absorbed as blurred or distorted Gaussian primitives. We propose LL-3DGS Reconstruction, a coupled geometric–photometric framework that estimates a canonical normally illuminated Gaussian scene while retaining raw low-light observations as explicit evidence. Its geometry front end preserves view identity and order, selects a raw or fixed enhanced stream for learned multi-view inference, and converts the estimated cameras, depth, and points into a density-aware Gaussian seed with optional Sim(3) co-registration. The Gaussian back end constrains one shared 3D field through a bounded luminance restoration path and a low-light re-degradation path, preventing independent per-view corrections from becoming inconsistent scene content. Dark-region weighting, reliability-aware supervision, staged activation, bounded densification, and scale–anisotropy constraints further regularize flexible Gaussian primitives against noise, illumination errors, and weak geometry. Finally, co-registered metric depth supplies surface geometry while the canonical Gaussian rendering supplies restored color for colored-point and TSDF fusion. With paired normal-light supervision, the method obtains 25.44 dB PSNR, 0.898 SSIM, and 0.247 LPIPS on five LOM scenes and averages 21.195 dB, 0.449 SSIM, and 0.523 LPIPS across eight LLRS scenes. A controlled LOM-sofa ablation shows that low-light consistency supports observation explainability, while four-scene TSDF outputs document the explicit reconstruction path without claiming geometric accuracy in the absence of 3D ground truth.
Monocular Gaussian SLAM must recover camera motion, surface structure, and appearance from an RGB sequence without metric depth input or benchmark geometry during reconstruction. This setting is challenged by scale-ambiguous predictions, spatially varying reliability, and the tendency of an unconstrained Gaussian map to absorb pose and depth errors into its geometric and appearance parameters. We present Mono3DGS-SLAM, a reliability-guided Gaussian–TSDF framework with a sequence-precalibrated prediction stage and an incremental mapping backend. Deterministic reliability gates select admissible observations, while one frozen scale jointly transports predicted depth, camera translation, and length-valued mapping parameters into an internally consistent canonical coordinate. A confidence-weighted colorized TSDF stores the persistent surface, base color, and visibility support. A capacity-bounded Gaussian layer is restricted to reliable residual regions and primarily restores appearance detail over the fused surface. On Replica, Mono3DGS-SLAM records 0.249 cm mean ATE in the evaluated runs and has lower ATE in five of eight scenes within the locally evaluated monocular group. With identical poses, depths, and views, the hybrid output reaches 30.678 dB PSNR in a paired comparison of outputs from the same hybrid runs. This comparison does not establish superiority over independently optimized single-representation systems. The evaluated full pipeline uses future observations and does not establish online or real-time SLAM performance. ScanNet and TUM-RGBD results further assess transfer to real monocular sequences.
Monocular Gaussian SLAM must recover camera motion, surface structure, and appearance from an RGB sequence without metric depth input or benchmark geometry during reconstruction. This setting is challenged by scale-ambiguous predictions, spatially varying reliability, and the tendency of an unconstrained Gaussian map to absorb pose and depth errors into its geometric and appearance parameters. We present Mono3DGS-SLAM, a reliability-guided Gaussian–TSDF framework with a sequence-precalibrated prediction stage and an incremental mapping backend. Deterministic reliability gates select admissible observations, while one frozen scale jointly transports predicted depth, camera translation, and length-valued mapping parameters into an internally consistent canonical coordinate. A confidence-weighted colorized TSDF stores the persistent surface, base color, and visibility support. A capacity-bounded Gaussian layer is restricted to reliable residual regions and primarily restores appearance detail over the fused surface. On Replica, Mono3DGS-SLAM achieves 0.249 cm mean ATE and is most accurate in five of eight scenes. With identical poses, depths, and views, the hybrid output reaches 30.678 dB PSNR, exceeding either component alone. ScanNet and TUM-RGBD results further assess transfer to real monocular sequences.