Vincent Fournier's Activities

[Contextcapture] benchmark scores

Oto — Fri, 13 Sep 2019 08:04:59 GMT

What are typical scores for benchmark in Contextcapture for different GPU cards? Does GPU RAM size influence much?

Benchmark scores can be checked in Contextcapture Settings->Configuration->Start Benchmark

My compiled list:

RX7900XTX - 982
RTX4090 - 551
RTX3090 have 446
Nvidia RTX 3080 benchmarks at 464 [375 Vulkan API]
RTX a6000 have 340
RTX 3070 have 310
RTX 2080 Ti - 303
NVIDIA Quadro RTX 3000 -138
2x AMD Radeon VII - 226[Vulkan API]
RTX 2080 - 215/100
Titan V (Volta) -?
Quadro P5200- 162
TITAN Xp -?
TITAN X (Pascal) (recommended)- ?
GTX TITAN X (GTX 1080 Ti) Maxwell - ?
GTX 1080 (now recommended) - (191/100) or in latest versions - 215/100
GTX 1070 - ?
GTX 980 Ti (Titan) (from CC recommended list) - exceeds it by far (150/100)
GTX 980 - just exceeds 100 (117/100)
GTX 1060 (recommended) - 99/100
NVIDIA GeForce GTX 980M - (75/100)
NVIDIA GeForce GTX 960M - (45/100)
GTX 780 Ti (from CC recommended list) - ??
Nvidia Tesla K40m - 57/100
NVIDIA Quadro K4000 - 24/100
NVIDIA Quadro K2000M - (20.3/100)
GTX 650 Ti - 24/100
GTX 260 - 12.6/100
GTX 730 - 10/100

Also there seems there is no linear correlation with benchmark scores regarding processing time. For example on same machine GTX 650 Ti processes sample model 5 hours while GTX 980 Ti processes it at 3 hours while benchmark scores differ 7 times performance is only 2 times.

General Contextcapture System Requirements:
Minimum Hardware
It requires at least 8 GB of RAM and NVIDIA or AMD graphics card, or Intel-integrated graphics processor compatible with OpenGL 3.2 with at least 1 GB of dedicated memory.

For specific Annotation jobs leveraging artificial intelligence, minimum GPU requirement are higher. A 4.0GB NVIDIA device with up-to-date drivers is the minimum.

Recommended Hardware
As of January 2020, the following configuration is recommended: a recent desktop computer running under Microsoft Windows 8/10 64-bit with 64 GB of RAM, an 8+core i9-CPU and an NVIDIA GeForce RTX 2080 Ti graphics card.
Input, working and output data should preferably be stored on fast storage devices (fast HDD, SSD, SAN). For file sharing, we recommend a >1-Gigabit Ethernet network.

Please post your scores!

Can we change the exposure of some pictures without going through the whole computing process all over again?

Vincent Fournier — Tue, 29 Nov 2022 09:20:47 GMT

On a large 7500 photos ortho photo project, I notice that there was some color issues on some parts of the ortho. I figured out later that this was due to a significant difference of exposures at some points where pictures were taken the day after the previous ones. Is there a way to replace the relevant pictures by the same ones with retouched exposure to correct the issue without going through the whole process (AT, reconstruction, production) again?

Using CC engines on a virtual computer in the cloud.

Vincent Fournier — Tue, 29 Nov 2022 09:26:52 GMT

I am considering using virtual computers in the cloud with a CC engine in order to speed some projects up. Technically, it seems possible using fr instance Appstream service from AWS. Did anyone consider this before or even better try it? I also wonder if you can use 2 engines with the same project, one being on your local computer and the other in the cloud (not Bentley's but a virtual computer).

Can I offset the base position of a 3MX model ?

Vincent Fournier — Wed, 24 Feb 2021 15:35:04 GMT

Hi, I want to combine two models, into a single 3MX combined model.

THis is easy to do by placing the two Data folders in a single scene floder and editing the .3mx file like this :

{"3mxVersion": 1,
"name":"Combined_model",
"description":"Scene generated by ContextCapture, copyright <a href='http://www.bentley.com' target='_blank'>Bentley</a>.",
"logo":"logo.png",
"sceneOptions":[{"navigationMode":"ORBIT"}],
"layers": [
{
"type":"meshPyramid",
"id":"mesh0",
"name":"Model_1",
"description":"Model generated by ContextCapture, copyright <a href='http://www.bentley.com' target='_blank'>Bentley</a>.",
"SRS": "Local:unit=meter",
"SRSOrigin": [0,0,0],
"root": "Data-1/Model_1.3mxb"
},
{
"type":"meshPyramid",
"id":"mesh0",
"name":"Model_2",
"description":"Model generated by ContextCapture, copyright <a href='http://www.bentley.com' target='_blank'>Bentley</a>.",
"SRS": "Local:unit=meter",
"SRSOrigin": [0,0,0],
"root": "Data-2/Model_2.3mxb"
}
]
}

However, when I do this the two models are one on top of the other, so that I want to offset one of them with respect to the other.
I tried changing one of the

"SRSOrigin":[0,0,0]statements

to something else (such as [0,2,0] ) , but this leads to an error (probably SRS Origin needs to be the same for the two mode).

Any idea if this is possible at all (including by editing the 3mxb root files in each of the Data folders) ?

Thanks

Philippe

PS of course, it would be a solution to produce these models with an offset, but that would request a new Reconstruction, and since I heavily edited both meshes, it's rather complex, if possible at all.

Import ContextCapture Model to lumenRT

Vincent Fournier — Mon, 20 Jan 2020 11:30:09 GMT

I want to animate a Large ContextCapture model in LumenRT.
I've tried importing in CCEditor, and using the LumenRT export in CCEditor, but all the LODs are lost, so that only the lowest resolution level are displayed in LumerRT.

Is there a tuto or method on how to do this export / import ?

Thanks

Philippe

Ask A Question I

Mon, 20 Jan 2020 01:56:42 GMT

Ask a question in a forum.