LimeLight V2+ 2023 Firmware

I thought I might break off and start a new thread about the LimeLight “Legacy” hardware and some quick settings about Frame Rate while detecting April Tags.
Images are captured at tag height and at a distance of 48" , I don’t know how everyone will use these vision targets but there is a good possibly we will be in close range when using the data
LLv2 running FW2023.0.0

90FPS is possible:
resolution 320 x 240 2x zoom
4 detector downscale (scans image at x division for marks) then uses full resolution for ID decoding.



~50 FPS with no Zoom for larger field of view
Resolution 640 x 480

Some New values published to the NT

Finally for the time:
The Json file published to the NT
{
“Results”: {
“Classifier”: ,
“Detector”: ,
“Fiducial”: [
{
“fID”: 1,
“fam”: “16H5C”,
“pts”: ,
“skew”: ,
“t6c_ts”: [
-0.07334052691587963,
0.13680749733243142,
-1.463909865659224,
-3.9483071041272388,
10.218297410557885,
-1.198558682500163
],
“t6r_fs”: [
5.765637369047267,
-2.8649785592556999,
0.0,
0.0,
0.0,
-10.29764802967124
],
“t6r_ts”: [
-0.07160948045435088,
0.33634026318256679,
-1.4774627259638234,
-3.9483071041272388,
10.218297410557885,
-1.198558682500163
],
“t6t_cs”: [
-0.18471846389556999,
-0.23505393839729905,
1.4414423173089686,
3.79582666980205,
-10.275387922188795,
0.5039979125491136
],
“t6t_rs”: [
-0.18471846389556999,
-0.4350539413775313,
1.4414423173089686,
3.79582666980205,
-10.275387922188795,
0.5039979125491136
],
“ta”: 0.01704251393675804,
“tx”: 9.45051383972168,
“txp”: 252.56143188476563,
“ty”: 15.73668384552002,
“typ”: 154.77963256835938
}
],
“Retro”: ,
“pID”: 0.0,
“tl”: 22.90166664123535,
“ts”: 2836469.465011,
“v”: 1
}
}

3 Likes

Can you list resolution as well?

NVM: I see it now

1 Like

I hoped the images would show any questions

2 Likes

I see it now. Mind linking this thread here: LimeLight V2+ 2023 Firmware - #2 by mdurrani834

Also, could you test 640x480 on decimate of one? (no zoom)

Appreciate this early testing.

2 Likes

Similar things here – curious how limelight fares in our standard benchmark test: Supported Hardware - PhotonVision Docs

Here is 25"

Resolution 320 x 240 60fps 3x zoom
detector downscale 2


does decimate = Detector Downscale?


Looks like Detector Downscale is important:
@2

@3

@4

Looks like each step gains ~10 fps. I question what disadvantage could each step creates?

2 Likes

Yes, decimate is the same as downscale (to my knowledge, Brandon may have things setup differently at the back end).

Decimate basically scales down the image to run detection at a lower resolution. This gives you better performance but limits your detection range significantly at each “step”.

1 Like

Resolution 1280 x 960 22fps
detector downscale 1
Pipeline FPS ~1

Resolution 1280 x 960 22fps
detector downscale 4
Pipeline FPS ~15

1 Like

Empirically I observing the opposite. Downscaling seems to improve the detection at distance.

Can you measure the max range at different downsamplings in a reasonably well-lit room? I can get ~25ft range at 640x480 without downsampling in PV, or 20ft of range with a Decimate of 2. It would be nice to basically correlate max detection distance vs. framerate for a particular resolution so we can make good choices with the settings.

2 Likes

I will not have the time to do thorough testing till this evening. Yes lighting will effect test.

1 Like

Out of curiosity how are you setup? Do you have a rio? Is your LL connected to your router or have your set your pc up for dhcp? Are you using the static ip?

Also any plans to try out the ML

I’m assuming your asking because of the other IP Config talks on the other thread. Your wright with ask though.

The above setup was just my PC and the LimeLight. No Rio No Radio. My PC was set static with 10.TE.AM.2 (to allow me to be the NetworkTable server)

I have Just completed a deeper dive into the IP config issue and it seems the 2023.0.1 firmware released late last night has resolved it. I too did have issues getting the IP set for the 2023.0.0 initial release but I have a few tools I used that I use everyday to get past this.

One thing to mention here is I did have to disconnect my machine from other networks. Once I did this and using the 2023.0.1 firmware the LimeLight Finder tool work flawlessly. I would advise either every time or if you having issues to disable other network adapters then use the LL Finder to discover and configure.

I am not privileged to the change log yet for the 2023.0.1 release but can only assume @Brandon_Hjelstrom has been busy getting the firmware out and the LL3’s ready to ship. Change Logs are a later problem.

2 Likes

I’m not following?

distance 5 ft
Resolution 640x480
detector downscale 4

distance 13 ft
Resolution 320x240 3x Zoom
detector downscale 2

2 Likes

After installing 2023.0.1 on both of our Limelight 2+es, one of them began registering this error with very poor performance:

The fan was operating normally. We then re-flashed 2022.3 to that device and weren’t able to access it in the Limelight discovery tool.

Our second Limelight appears to work as expected with good performance.

We’re going to take apart the “CHECK FAN” Limelight to see if the heatsink was dislodged or other thermal/hardware anomalies are present, but @Brandon_Hjelstrom any suggestions here?

1 Like

Perfect, this is exactly what I wanted. Thanks!

Sorry the ML was for machine learning. I left recognition beyond just April tags

Is there an optimal height for reading April tags and estimating pose this year?

I’ve heard head-on causes the most pose ambiguity.

I’m no expert but won’t placing it at an odd angle eliminate the head on ambiguity problem as it will almost never be aligned to a tag?