Add per-dataset and overall summary rows to benchmark results table (#4361)
## Summary
- New `aggregate_scene_metrics` helper computes `__avg__` (mean of
per-scene metrics) and `__all__` (recomputed from pooled raw errors);
applied at category, dataset, and overall levels.
- `Metrics` now carries `errors` and `position_accuracy_gt` so
higher-level pooled aggregates can be reconstructed.
- `create_result_table` emits a `###dataset###` block per dataset with
>1 category and a final `###overall###` block when there's >1 dataset.
- Side fix: per-category `__all__` rows previously reported dataset-wide
image/component totals; now correctly per-category.
Looks like this now:
```
======scenes======= ======AUC @ X deg (%)====== ===images=== =components=
0.5 1.0 5.0 10.0 reg all num largest
=============================imc2023=haiper==============================
bike 4.17 22.66 62.63 68.46 15 15 1 15
chairs 12.27 43.38 88.45 94.23 16 16 1 16
fountain 57.63 78.59 95.72 97.86 23 23 1 23
-------------------------------------------------------------------------
overall 34.40 57.38 86.60 90.47 54 54 3 54
average 24.69 48.21 82.27 86.85 18 18 1 18
==============================imc2023=urban==============================
kyiv-puppet-theater 1.16 6.50 31.65 43.52 25 26 1 25
-------------------------------------------------------------------------
overall 1.16 6.50 31.65 43.52 25 26 1 25
average 1.16 6.50 31.65 43.52 25 26 1 25
============================imc2023=heritage=============================
cyprus 44.92 68.51 88.37 90.85 29 30 1 29
dioscuri 45.54 63.51 87.73 92.71 173 174 1 173
wall 50.38 74.43 94.89 97.44 43 43 1 43
-------------------------------------------------------------------------
overall 45.79 64.24 88.14 92.92 245 247 3 245
average 46.95 68.82 90.33 93.67 82 82 1 82
==========================imc2023=phototourism===========================
piazza_san_marco 15.17 39.62 80.60 89.39 68 68 1 68
sagrada_familia 22.94 37.94 59.00 62.95 90 90 1 90
st_pauls_cathedral 21.35 41.90 74.29 82.00 142 142 1 142
-------------------------------------------------------------------------
overall 20.87 40.61 71.41 78.35 300 300 3 300
average 19.82 39.82 71.30 78.11 100 100 1 100
#################################imc2023#################################
overall 33.07 52.08 79.43 85.31 624 627 10 624
average 27.55 47.70 76.33 81.94 62 63 1 62
===============================eth3d=dslr================================
door 24.19 46.95 84.62 92.55 7 7 1 7
lounge 21.20 27.26 32.12 32.73 6 10 1 6
statue 94.47 97.23 99.45 99.72 11 11 1 11
-------------------------------------------------------------------------
overall 54.86 62.34 71.81 73.55 24 28 3 24
average 46.62 57.15 72.06 75.00 8 9 1 8
================================eth3d=rig================================
storage_room 7.49 13.42 21.97 23.43 398 796 1 398
storage_room_2 1.68 5.09 17.30 20.75 416 832 1 416
terrains 0.86 4.40 18.20 21.31 330 660 1 330
-------------------------------------------------------------------------
overall 3.57 7.91 19.21 21.85 1144 2288 3 1144
average 3.34 7.64 19.16 21.83 381 763 1 381
##################################eth3d##################################
overall 3.57 7.92 19.21 21.86 1168 2316 6 1168
average 24.98 32.39 45.61 48.41 195 386 1 195
#################################overall#################################
overall 4.65 9.54 21.42 24.19 1792 2943 16 1792
average 26.59 41.96 64.81 69.37 112 184 1 112
``` J
Johannes Schönberger committed
ffb5404e7bbfb9c718726f16718d7264582fe718
Parent: 81e6145
Committed by GitHub <noreply@github.com>
on 4/24/2026, 6:45:41 PM