| CARVIEW |
Select Language
HTTP/2 301
server: GitHub.com
content-type: text/html
location: https://brabbitdousha.github.io/MIRReS/
access-control-allow-origin: *
strict-transport-security: max-age=31556952
expires: Sun, 28 Dec 2025 22:41:25 GMT
cache-control: max-age=600
x-proxy-cache: MISS
x-github-request-id: 5A03:123DE:809700:905EFE:6951AFBC
accept-ranges: bytes
age: 0
date: Sun, 28 Dec 2025 22:31:25 GMT
via: 1.1 varnish
x-served-by: cache-bom-vanm7210036-BOM
x-cache: MISS
x-cache-hits: 0
x-timer: S1766961085.333798,VS0,VE211
vary: Accept-Encoding
x-fastly-request-id: 0b37b1b6d54d797b8d0e4b167fe6b0f2a852ec89
content-length: 162
HTTP/2 200
server: GitHub.com
content-type: text/html; charset=utf-8
last-modified: Wed, 24 Sep 2025 17:53:38 GMT
access-control-allow-origin: *
strict-transport-security: max-age=31556952
etag: W/"68d43022-4c75"
expires: Sun, 28 Dec 2025 22:41:25 GMT
cache-control: max-age=600
content-encoding: gzip
x-proxy-cache: MISS
x-github-request-id: 0DD9:3157C7:7FD692:8F9DD1:6951AFBD
accept-ranges: bytes
age: 0
date: Sun, 28 Dec 2025 22:31:25 GMT
via: 1.1 varnish
x-served-by: cache-bom-vanm7210036-BOM
x-cache: MISS
x-cache-hits: 0
x-timer: S1766961086.558443,VS0,VE210
vary: Accept-Encoding
x-fastly-request-id: 417eaf9dd12331b1d7d8a3ca09e9f2851b57cd36
content-length: 5319
MIRReS: Inverse Rendering using Multi-Bounce Path
Tracing and Reservoir Sampling
ICLR 2025
Note that in terms of novel view synthesis, our method demonstrates similar performance compared to TensoIR and GS-IR. This is because these baselines tend to overfit the training dataset and bake the complex lighting effects (eg. shadows) within the materials, as qualitatively analyzed above. Despite this, we can produce considerably better material estimation and relighting results.
MIRReS: Inverse Rendering using Multi-Bounce Path Tracing and Reservoir Sampling
ICLR 2025
- Yuxin Dai
- Qi Wang
- Jingsen Zhu
- Dianbing Xi
- Yuchi Huo
- Chen Qian
- Ying He
Abstract
We present MIRReS, a novel two-stage inverse rendering framework that jointly reconstructs and optimizes explicit geometry, materials, and lighting from multi-view images. Unlike previous methods that rely on implicit irradiance fields or oversimplified path tracing algorithms, our method first extracts an explicit triangular mesh in the initial stage. Subsequently, it employs a more realistic physically-based inverse rendering model in the second stage, utilizing multi-bounce path tracing and Monte Carlo integration. By leveraging multi-bounce path tracing, our method not only effectively estimates indirect illumination--including self-shadowing and internal reflections--but also enhances the intrinsic decomposition of shape, material, and lighting. Moreover, we incorporate reservoir sampling into our framework to address the noise in Monte Carlo integration, enhancing convergence and facilitating gradient-based optimization with low sample counts. Through both qualitative and quantitative assessments across various scenarios, especially those with complex shadows, we demonstrate that our method achieves state-of-the-art performance in decomposition results. Additionally, our optimized explicit geometry supports further applications in scene editing, relighting, and material editing, compatible with modern graphics engines and CAD software.
Quantitative comparison
Note that in terms of novel view synthesis, our method demonstrates similar performance compared to TensoIR and GS-IR. This is because these baselines tend to overfit the training dataset and bake the complex lighting effects (eg. shadows) within the materials, as qualitatively analyzed above. Despite this, we can produce considerably better material estimation and relighting results.
Albedo results and relighting results
Drag the slider to compare!
Albedo results
Normal results
visibility results
Pipeline
This video will explain our motivation and each stage of the pipeline, and showcase our results.
Overview
Results on TensoIR-Synthetic Dataset
Relighting results
Benefits
Scene editing and material editing
ReSTIR result under 1 spp and indirect shading effects
BibTeX
@inproceedings{dai2025inverse,
title = {Inverse Rendering for Shape, Light, and Material Decomposition using Multi-Bounce Path Tracing and Reservoir Sampling},
author = {Yuxin Dai and Qi Wang and Jingsen Zhu and Dianbing Xi and Yuchi Huo and Chen Qian and Ying He},
booktitle = {The Thirteenth International Conference on Learning Representations},
year = {2025},
url = {https://openreview.net/forum?id=KEXoZxTwbr}
}