haian-jin/tensoir

How haian-jin/tensoir works

TensoIR is a PyTorch inverse-rendering system for learning scene geometry, materials, and lighting from multi-view images. Its main workflow loads scene images and lighting, trains a tensor-factorized scene representation, then renders learned-lighting views or relights the scene under unseen environment maps; it can also export a mesh. The README documents single-light, rotated multi-light, general multi-light, and optional NeRF-Synthetic training modes. Internal links are limited to relationships supported by the README or source-index references; several implementation details were not sampled.

Main entry points: renderer.py Renderer · tensoRF_rotated_lights.py Rotated Tensor Field · opt.py Run Configuration · relight_importance.py Relight Sampling

Components 25

Scene Inputs

Scene Images
Environment Maps
Run Configuration opt.py
Camera Rays dataLoader/ray_utils.py
Blender Dataset dataLoader/blender.py
Single-Light Dataset dataLoader/tensoIR_simple.py
Rotated-Light Dataset dataLoader/tensoIR_rotation_setting.py
General-Light Dataset dataLoader/tensoIR_general_multi_lights.py

Scene Learning

Single-Light Training train_tensoIR.py
NeRF-Synthetic Training train_tensoIR_simple.py
Rotated-Light Training train_tensoIR_rotated_multi_lights.py
General-Light Training train_tensoIR_general_multi_lights.py
Rotated Tensor Field models/tensoRF_rotated_lights.py
General Tensor Field models/tensoRF_general_multi_lights.py
Tensor Field Base models/tensorBase_rotated_lights.py
General Field Base models/tensorBase_general_multi_lights.py
Initial Tensor Field models/tensoRF_init.py
SH Lighting Basis models/sh.py

Rendering and Relighting

Relighting Utilities models/relight_utils.py
Renderer renderer.py
Relight Sampling scripts/relight_importance.py

Analysis and Outputs

Mesh Export scripts/export_mesh.py
Image Metrics utils.py
Rendered Results

Other components

Researcher

Connections 37

Researcher → Run Configuration: sets options README.md
Scene Images → Single-Light Dataset: loads images
Environment Maps → General-Light Dataset: loads lighting dataLoader/tensoIR_general_multi_lights.py
Camera Rays → Single-Light Dataset: generates rays dataLoader/tensoIR_simple.py
Camera Rays → Rotated-Light Dataset: generates rays dataLoader/tensoIR_rotation_setting.py
Camera Rays → General-Light Dataset: generates rays dataLoader/tensoIR_general_multi_lights.py
Run Configuration → Single-Light Training: configures run README.md
Run Configuration → NeRF-Synthetic Training: configures run README.md
Run Configuration → Rotated-Light Training: configures run README.md
Run Configuration → General-Light Training: configures run README.md
Single-Light Dataset → Single-Light Training: provides samples
Blender Dataset → NeRF-Synthetic Training: provides samples
Rotated-Light Dataset → Rotated-Light Training: provides samples
General-Light Dataset → General-Light Training: provides samples
Single-Light Training → Rotated Tensor Field: trains model train_tensoIR.py
NeRF-Synthetic Training → Rotated Tensor Field: trains model train_tensoIR_simple.py
Rotated-Light Training → Rotated Tensor Field: trains model train_tensoIR_rotated_multi_lights.py
General-Light Training → General Tensor Field: trains model train_tensoIR_general_multi_lights.py
Rotated Tensor Field → Tensor Field Base: extends field models/tensoRF_rotated_lights.py
General Tensor Field → General Field Base: extends field models/tensoRF_general_multi_lights.py
Tensor Field Base → SH Lighting Basis: evaluates basis models/tensorBase_rotated_lights.py
General Field Base → SH Lighting Basis: evaluates basis models/tensorBase_general_multi_lights.py
Relighting Utilities → Initial Tensor Field: uses alpha conversion models/relight_utils.py
Renderer → Relighting Utilities: renders BRDF renderer.py
Single-Light Training → Renderer: renders tests README.md
NeRF-Synthetic Training → Renderer: renders tests README.md
Rotated-Light Training → Renderer: renders tests README.md
General-Light Training → Renderer: renders tests README.md
Researcher → Relight Sampling: requests relighting README.md
Relight Sampling → Rotated Tensor Field: loads checkpoint scripts/relight_importance.py
Relight Sampling → Renderer: renders relighting scripts/relight_importance.py
Relight Sampling → Environment Maps: samples unseen light README.md
Renderer → Image Metrics: computes metrics renderer.py
Renderer → Rendered Results: saves renders README.md
Researcher → Mesh Export: requests extraction README.md
Mesh Export → Rotated Tensor Field: loads checkpoint scripts/export_mesh.py
Mesh Export → Rendered Results: writes mesh README.md

More Python diagrams: ymahlau/fdtdx · blackhatethicalhacking/secretopt1c · brekkylab/backlot · livekit-examples/agent-starter-python · mit-spark/buffer-x

Drawn by GitDiagram on September 29, 2026 from the repository's file tree, README and source files. Read as Markdown ·