experiment stringclasses 1
value | arm stringclasses 7
values | step int64 0 30k | mean_psnr_db_online float64 8.99 19.8 | mean_ssim_online float64 0.58 0.74 | mean_lpips_online float64 0.49 0.58 | total_material_hole_pixels int64 0 16.1k | mean_rpe_translation_before float64 1 1 | mean_rpe_translation_after float64 0.97 1 | mean_rpe_rotation_deg_before float64 3.3 3.3 | mean_rpe_rotation_deg_after float64 2.57 3.3 | mean_input_blur_psnr_db float64 21.3 21.3 | mean_alpha_valid float64 0.97 1 | mean_coverage_alpha_gt_005 float64 1 1 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
diag356 | coupled_lr2e6 | 0 | 19.849442 | 0.738408 | 0.492726 | 0 | 1.003149 | 1.003149 | 3.303192 | 3.303192 | 21.275005 | 0.996482 | 1 |
diag356 | coupled_lr2e6 | 10,000 | 9.203089 | 0.599992 | 0.557247 | 8,747 | 1.003149 | 0.98735 | 3.303192 | 2.74746 | 21.275005 | 0.987806 | 0.998907 |
diag356 | coupled_lr2e6 | 30,000 | 9.718208 | 0.593683 | 0.56809 | 14,721 | 1.003149 | 0.977526 | 3.303192 | 2.570576 | 21.275005 | 0.977338 | 0.998043 |
diag356 | coupled_lr2e7 | 0 | 19.849442 | 0.738408 | 0.492726 | 0 | 1.003149 | 1.003149 | 3.303192 | 3.303192 | 21.275005 | 0.996482 | 1 |
diag356 | coupled_lr2e7 | 10,000 | 8.994342 | 0.605752 | 0.55489 | 0 | 1.003149 | 0.998148 | 3.303192 | 3.049638 | 21.275005 | 0.993178 | 0.999999 |
diag356 | coupled_lr2e7 | 30,000 | 9.776631 | 0.615314 | 0.550849 | 5,736 | 1.003149 | 0.993389 | 3.303192 | 2.772792 | 21.275005 | 0.985098 | 0.999187 |
diag356 | isolated_lr2e6 | 0 | 19.849442 | 0.738408 | 0.492726 | 0 | 1.003149 | 1.003149 | 3.303192 | 3.303192 | 21.275005 | 0.996482 | 1 |
diag356 | isolated_lr2e6 | 10,000 | 9.235501 | 0.60051 | 0.555563 | 8,778 | 1.003149 | 0.987374 | 3.303192 | 2.744582 | 21.275005 | 0.986661 | 0.998898 |
diag356 | isolated_lr2e6 | 30,000 | 9.711658 | 0.591879 | 0.569168 | 14,922 | 1.003149 | 0.97753 | 3.303192 | 2.571156 | 21.275005 | 0.975696 | 0.99802 |
diag356 | isolated_lr2e7 | 0 | 19.849442 | 0.738408 | 0.492726 | 0 | 1.003149 | 1.003149 | 3.303192 | 3.303192 | 21.275005 | 0.996482 | 1 |
diag356 | isolated_lr2e7 | 10,000 | 8.98945 | 0.605407 | 0.55462 | 0 | 1.003149 | 0.998148 | 3.303192 | 3.049086 | 21.275005 | 0.9929 | 0.999999 |
diag356 | isolated_lr2e7 | 30,000 | 9.719669 | 0.613407 | 0.550983 | 5,691 | 1.003149 | 0.993406 | 3.303192 | 2.772568 | 21.275005 | 0.984365 | 0.999192 |
diag356 | pose_lr_1e6 | 0 | 19.849442 | 0.738408 | 0.492726 | 0 | 1.003149 | 1.003149 | 3.303192 | 3.303192 | 21.275005 | 0.996482 | 1 |
diag356 | pose_lr_1e6 | 10,000 | 9.205919 | 0.603195 | 0.555361 | 5,049 | 1.003149 | 0.991158 | 3.303192 | 2.811657 | 21.275005 | 0.990953 | 0.999252 |
diag356 | pose_lr_1e6 | 30,000 | 9.264948 | 0.584715 | 0.575566 | 10,261 | 1.003149 | 0.984516 | 3.303192 | 2.632874 | 21.275005 | 0.982918 | 0.998521 |
diag356 | pose_lr_2e6 | 0 | 19.849442 | 0.738408 | 0.492726 | 0 | 1.003149 | 1.003149 | 3.303192 | 3.303192 | 21.275005 | 0.996482 | 1 |
diag356 | pose_lr_2e6 | 10,000 | 9.277591 | 0.599626 | 0.555361 | 9,266 | 1.003149 | 0.985769 | 3.303192 | 2.740384 | 21.275005 | 0.986177 | 0.998827 |
diag356 | pose_lr_2e6 | 30,000 | 9.669246 | 0.589299 | 0.568763 | 16,071 | 1.003149 | 0.972447 | 3.303192 | 2.565789 | 21.275005 | 0.9729 | 0.997838 |
diag356 | pose_lr_2e7 | 0 | 19.849442 | 0.738408 | 0.492726 | 0 | 1.003149 | 1.003149 | 3.303192 | 3.303192 | 21.275005 | 0.996482 | 1 |
diag356 | pose_lr_2e7 | 10,000 | 9.067834 | 0.606875 | 0.554418 | 0 | 1.003149 | 0.998137 | 3.303192 | 3.048696 | 21.275005 | 0.992247 | 0.999998 |
diag356 | pose_lr_2e7 | 30,000 | 9.724838 | 0.613307 | 0.554553 | 5,708 | 1.003149 | 0.993362 | 3.303192 | 2.77222 | 21.275005 | 0.984787 | 0.999192 |
AnyDeblur3D Replica completed eval archive
This private dataset preserves visualizable, complete-terminal Replica results before selected DAS6 scratch roots are removed.
per_target_eval: 609 rows = 7 completed arms x 3 retained steps x 29 targets.arm_step_summary: 21 aggregate arm-step rows.completed/: original complete effect JSON, effect panels, arm terminals, RPE traces, status files, and compact trajectory evidence for diag356/357.failures/: structural control, status, logs, receipts, and small checkpoint evidence for the frozen s16 and s18 teacher failures. These are engineering failures, not scientific results.
The completed experiments use one seed. PSNR differences below 0.5 dB are treated as practical near-ties; no statistical-significance claim is made. PENDING/RUNNING/engineering receipts are never promoted to scientific PASS.
Large regenerable teacher caches, CUDA-extension caches, and historical dense arrays/checkpoints are intentionally not included in the visualizable subset. File-level SHA-256 manifests accompany each retained archive tree.
- Downloads last month
- 67