gerw commited on
Commit
f6351bb
·
verified ·
1 Parent(s): e755bf7

Add dataset card

Browse files
Files changed (1) hide show
  1. README.md +51 -0
README.md ADDED
@@ -0,0 +1,51 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: mit
3
+ pretty_name: DiffReg-PBIR — Stanford-ORB submission
4
+ tags:
5
+ - inverse-rendering
6
+ - stanford-orb
7
+ - relighting
8
+ task_categories:
9
+ - image-to-3d
10
+ ---
11
+
12
+ # DiffReg-PBIR — Stanford-ORB benchmark submission
13
+
14
+ Predicted outputs and evaluation scores for **DiffReg-PBIR** on the
15
+ [Stanford-ORB](https://stanfordorb.github.io/) inverse-rendering benchmark
16
+ (full 42-scene split).
17
+
18
+ **Paper:** *Diffusion-Based Material Regularization for Physics-Based Inverse Rendering* (ECCV 2026)
19
+ - Project page: https://gerwang.github.io/diffusion-regularized-inverse-rendering/
20
+ - arXiv: https://arxiv.org/abs/2606.31065
21
+
22
+ ## Contents (`submission_DiffReg-PBIR.zip`, ~6.8 GB uncompressed)
23
+
24
+ ```
25
+ submission_DiffReg-PBIR/
26
+ predictions/<scene>/
27
+ mesh.obj # reconstructed mesh (shape task)
28
+ view/*.exr # HDR novel-view renders
29
+ light/*.exr # HDR novel-scene relit renders
30
+ geometry/*_image.zbuf.exr, *_image.normal.exr # Z-depth + camera-space normals
31
+ material/*_albedo.exr # albedo
32
+ test_results.json # paths in the official examples/test/mymethod.json schema
33
+ # output_* relative to package root
34
+ # target_* relative to orb_data/ (= blender_HDR/ + ground_truth/)
35
+ scores.json # evaluation scores (upstream scripts.test output)
36
+ ```
37
+
38
+ ## Results (full split) vs. previous leaderboard best (Neural-PBIR)
39
+
40
+ | Metric | DiffReg-PBIR | Neural-PBIR |
41
+ |---|---|---|
42
+ | Relight PSNR-H ↑ | **27.22** | 26.01 |
43
+ | Relight PSNR-L ↑ | **34.98** | 33.26 |
44
+ | Relight SSIM ↑ | **0.981** | 0.979 |
45
+ | Relight LPIPS ↓ | **0.021** | 0.023 |
46
+ | Novel-view PSNR-H ↑ | **29.58** | 28.82 |
47
+ | Normal (angular) ↓ | **0.014** | 0.06 |
48
+
49
+ Reproduce: point `orb_data/` at the official Stanford-ORB data root and run
50
+ `python scripts/test.py -i test_results.json -o scores.json -s full` from the
51
+ upstream repo. Prediction EXRs are 3-channel RGB, HDR only (the eval derives LDR).