T3B — Depth estimation — photometric shift¶
Paper evidence: main.pdf · Block findings
Lemma: D3 · Stack: pytorch
Nuisance key: augmentation
Production change: Lighting/texture shift; depth target meaning unchanged.
Notebook (Run All, built-in demo): t03b-depth-augmentation.ipynb
pip install matching-pmh torch
# Open the notebook and Run All
What this task achieved (headline)¶
E1_aniso beats E1 on hard photometric depth stress (combined_hard AbsRel 0.2152 vs 0.2191).
| baseline | E1 | E1_aniso |
|---|---|---|
| 0.2033 AbsRel | 0.1951 | 0.2152 (photo hard) |
Paper preset: t3b_depth_d3 · from pmh.benchmark.presets import get_preset
Subtasks (paper)¶
NYU depth pipeline (train + calibrate)¶
E1_aniso wins photometric hard stress.
Preset: t3b_depth_d3
Photometric subspace calibration¶
Aug-delta Gram rank 32.
Preset: t3b_depth_d3
E1_wrong control (random W)¶
Strengthening / falsification arm.
Run with matching-pmh¶
from pmh import PMHTrainer, evaluate_robust_fit
# nuisance="augmentation"
Do not use PMH when¶
Different depth semantics or scale definition at deploy.
Replace demo data with yours¶
Swap demo loaders for your train_loader, source_batches, target_batches, and deploy holdout. Hook the backbone before your task head.