Raise minimum images for calibration to 100#2437
Conversation
58c5d67 to
c85b45f
Compare
Is this true? |
I've done like 200 in 10 minutes or so, assuming you have someone else to click the button that's about a picture every three seconds. For only 100, that's a picture every six seconds. |
|
I quite comfortably did ~250 image calibrations in the pits at worlds in 2025 between matches so I don't see 100 being a huge barrier |
|
Some sort of bypass option may be useful |
439f5fc to
b8e9e4f
Compare
c85b45f to
b615e69
Compare
49f300e to
8c974b1
Compare
|
I'd say 50 is a nice compromise. For our team at least, calibrations take longer to make sure we get proper images, and having it raised to 100 will improve frustrations for quite a few teams (including ours) that calibrate slower and need to quickly calibrate. To me, being limited by the OS to collect 100 images rather than 12 doesn't seem like a good change for usability and quick setup. That being said though, maybe we could have a limit of 50 and then a small banner in the calibration menu somewhere that says 100 images is strongly recommended... |
Have you read the linked document? We're choosing 100 images because that's the empirically defined point of diminishing returns, but if you can provide evidence that you'll be able to get sufficient results with 50 images all the time, we might be willing to reconsider. Something to keep in mind here is that we want to help teams fall into success, and that means forcing them to get a good calibration. If we give them a recommendation, many teams won't follow it. |
|
We're also going to try exploring some new calibration methods/guides over the offseason, TBD what that looks like |
1248520 to
64a6641
Compare
|
Current issues with bypass:
|
64a6641 to
0896631
Compare
|
I hath discovered that this is quite cursed. Basically we send stuff to the backend for calib using this guy: photonvision/photon-client/src/types/WebsocketDataTypes.ts Lines 84 to 94 in 363b827 |
|
Hi. I'm the mrcal; just stumbled on this. The numbers on the choreography page are scenario-specific, and you'll get something different with different cameras (fov, resolution) and different geometry and a different chessboard. Sometimes 50 images is more than enough, and sometimes not. mrcal gives you uncertainty maps, so you can get 50 (or whatever) images, look at the uncertainty, and use that to decide if you need more images or not. I don't know if your tooling makes that difficult, but that would be my recommendation. |
109c5b7 to
ffd386d
Compare
| :label-cols="6" | ||
| :switch-cols="6" |
There was a problem hiding this comment.
has this been fixed?
| }; | ||
|
|
||
| const drawAllSnapshots = ref(true); | ||
| // We really gotta fix our typing system |
There was a problem hiding this comment.
is this a legit todo? in that case cut an issue with more details
We can see on https://mrcal.secretsauce.net/docs-2.0/tour-choreography.html that more images appear to always be better, but diminishing returns kick in around 100 images. This shouldn't be too time consuming as a minimum and should get users the best effort to quality ratio so I think to makes sense as the default.
Co-authored-by: Matt Morley <matthew.morley.ca@gmail.com>
3517f61 to
d97a532
Compare



Description
We can see on
https://mrcal.secretsauce.net/docs-2.0/tour-choreography.html that more images appear to always be better, but diminishing returns kick in around 100 images. This shouldn't be too time consuming as a minimum and should get users the best effort to quality ratio so I think to makes sense as the default.
This PR also adds a bypass method for developers looking to test that will lower the limit to 10 images.
Meta
Merge checklist: