OceanLab AI / Command Center
Idle
Job ID: Pending
MCMC (Recommended)
Standard Splatfacto
Job #--
MCMC Mode
0 / 30,000 iter Refine Stop: 25,000
Points (Gaussians) SSIM
0 --
Loss (L1) VRAM
-- --
Data Processing
Target Frame Count (30 ~ Unlimited)
Video Input Only: 設定抽取張數。設為超大值 (如 10000) 即為全數抽取。
Strategy & Regularization
Total Iterations30,000
Stop Split At (Refine Stop)25,000
MCMC: 此為「冷卻期」開始點，建議設為總步數 80%。
Warmup
Refine Interval
Random Background
隨機背景色 (黑/白)，強迫清除浮游點。
Antialiased
反鋸齒渲染。開啟較平滑，關閉較銳利。
MCMC Dynamic Tuning
Noise LR (雜訊學習率)1,000,000
Opacity Reg (透明度約束)0.01
Scale Reg (尺寸約束)0.01
Cap Max (球數上限)1.0M
Reset Alpha Interval (重置週期)3,000 iter
每隔 N 步重置透明度，有助於清除霧氣。
Camera Optimizer
相機姿態微調
Expert Safety Controls
Cull Alpha (剔除閾值)0.001
SSIM Weight (結構權重)0.4
Downscale Factor (降採樣)2x
Export Settings
Snapshot/Export Threshold0.05
影響預覽快照與最終匯出的點雲密度。
Ultimates (終極畫質)
Use Scale Regularization
Use Bilateral Grid
Ready to Initialize
[System] Command Center Ready.
Training Center
Scene Name
Task Objective
Full Training
SfM + 3DGS Training (End-to-End)
Pose Estimation
Structure from Motion Only (No 3DGS)
Data Source
Upload Data
Open Existing
Advanced Pipeline
SfM Method
COLMAP
Standard
GLOMAP
Fast Global
Reality Scan
CSV + PLY
Existing
Load & Skip
Camera Optimizer
Continue Cull (Default)
Feature Extractor
Matcher
Max Keypoints2048