How kzhang2/aoneus works
AONeuS reconstructs 3D surfaces by jointly fitting imaging-sonar and optical observations with neural rendering. The documented demo starts from paired sensor data and configurations, runs the fusion training entry point (which can initialize the NeuS optical runner), then produces mesh and view-synthesis results. The sampled code supports the model, renderer, and runner relationships shown; README capabilities fill in the documented result branches. Detailed data-to-runner wiring is not fully visible in the excerpts.
Main entry points: run_sdf.py Fusion runner · exp_runner.py NeuS runner · fields.py Fusion fields · renderer.py Sonar renderer
Components 17
Run and configure
- Fusion runner
run_sdf.py - Sonar config
confs/turtle_reduced_baseline_0.6x_joint_sonar.conf - Optical config
confs/turtle_reduced_baseline_0.6x_joint_rgb.conf - NeuS runner
NeuS/exp_runner.py
Sensor inputs
- Sonar observations
data/reduced_baseline_0.6x_sonar - RGB observations
data/reduced_baseline_0.6x_rgb - Sonar data loader
load_data.py - Optical dataset
NeuS/models/dataset.py
Neural representations
- Fusion fields
models/fields.py - NeuS fields
NeuS/models/fields.py - Position encoding
models/embedder.py
Rendering and results
- Sonar renderer
models/renderer.py - Optical renderer
NeuS/models/renderer.py - Surface mesh
models/renderer.py - Synthesized views
NeuS/models/renderer.py
Other components
- Researcher
- Weights & Biases
Connections 16
- Researcher → Fusion runner: starts demo
- Sonar config → Fusion runner: configures
- Optical config → Fusion runner: configures
- Sonar observations → Sonar data loader: reads
- RGB observations → Optical dataset: loads
- Fusion runner → NeuS runner: initializes
- Fusion runner → Fusion fields: trains
- NeuS runner → NeuS fields: constructs
- Fusion fields → Position encoding: encodes inputs
- Fusion runner → Sonar renderer: renders sonar
- NeuS runner → Optical renderer: uses
- Sonar renderer → Fusion fields: queries fields
- Optical renderer → NeuS fields: queries fields
- Fusion runner → Surface mesh: exports mesh
- NeuS runner → Synthesized views: renders views
- Fusion runner → Weights & Biases: logs metrics