Contributor
I think this new hyper-parameter is a more sensible choice.
…h/min-model-size
Contributor
This has a slight risk of affecting the benchmarking, when many small models are saved and we reach max_num_models more quickly?
On the other hand, min_model_size is calculated with respect to number of images, but tested against number of registered frames. Is this a bug?
Contributor Author
This has a slight risk of affecting the benchmarking, when many small models are saved and we reach max_num_models more quickly?
Not sure if this is an issue. There can only be 2 models with at least half the number of images. So, unless one configured max_num_models=1, this shouldn't cause an issue?
On the other hand, min_model_size is calculated with respect to number of images, but tested against number of registered frames. Is this a bug?
This seems to be a bug. I fixed it.
Contributor
Not sure if this is an issue. There can only be 2 models with at least half the number of images. So, unless one configured max_num_models=1, this shouldn't cause an issue?
From my understanding, in the mode with multiple models, the second model does not exclude the images from the first model.
This seems to be a bug. I fixed it.
This seems to be the case not only in Python but also in C++, and in two places.
Contributor Author
From my understanding, in the mode with multiple models, the second model does not exclude the images from the first model.
There can be overlap. I am still convinced that if we reconstruct more than ~50% of the images, then we should keep the model. There is very little chance that when starting from another image pair, we would reconstruct more.
This seems to be the case not only in Python but also in C++, and in two places.
Yes, and it should be fixed in both.
Contributor
Contributor Author
OK, I went the extra mile to make sure there is no regression with this change. A: master, B: this branch
I20260112 14:02:13.463241 20457612 compare.py:main:58] Results A - B:
=====scenes===== ======AUC @ X deg (%)====== ===images=== =components=
0.5 1.0 5.0 10.0 reg all num largest
==============================eth3d=dslr==============================
botanical_garden -0.01 -0.00 -0.00 -0.00 0 0 0 0
boulders -0.00 -0.00 -0.00 -0.00 0 0 0 0
bridge 0.00 0.00 0.00 0.00 0 0 0 0
courtyard -0.02 -0.87 -0.45 -0.23 0 0 0 0
delivery_area -0.00 -0.00 -0.00 -0.00 0 0 0 0
door 0.00 0.00 0.00 -0.00 0 0 0 0
electro -0.00 -0.00 -0.00 -0.00 0 0 0 0
exhibition_hall -0.20 0.08 -0.06 0.22 0 0 0 0
facade -0.01 -0.01 -0.00 -0.00 0 0 0 0
kicker -1.07 -0.49 -2.04 -3.77 0 0 0 0
lecture_room -0.00 -0.00 -0.00 -0.00 0 0 0 0
living_room -0.00 -0.00 -0.00 -0.00 0 0 0 0
lounge -23.87 -28.62 -32.39 -32.86 -5 0 -1 -4
meadow 0.00 0.00 0.00 0.00 0 0 0 0
observatory -1.42 -1.28 -0.35 -0.18 0 0 0 0
office 0.00 0.00 0.00 0.00 0 0 0 0
old_computer -0.00 -0.00 -0.00 -0.00 0 0 0 0
pipes 4.23 3.16 0.68 0.34 0 0 0 0
playground -0.00 -0.00 -0.00 -0.00 0 0 0 0
relief -0.00 -0.00 -0.00 -0.00 0 0 0 0
relief_2 -0.00 -0.00 -0.00 -0.00 0 0 0 0
statue -0.00 -0.00 -0.00 -0.00 0 0 0 0
terrace -0.00 -0.00 -0.00 -0.00 0 0 0 0
terrace_2 0.00 0.00 0.00 0.00 0 0 0 0
terrains 1.11 1.15 0.24 0.12 0 0 0 0
----------------------------------------------------------------------
overall -0.05 -0.05 -0.12 -0.13 -5 0 -1 -4
----------------------------------------------------------------------
average -0.85 -1.08 -1.38 -1.45 0 0 0 0
…h/min-model-size
Contributor
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thank you!
…h/min-model-size