Yet again out of Memory

[FoodMan]

hello Dmitry..

If I understand you right.. one should always have a bounding box as tight as possible...?

f/

[Dmitry Semyonov]

Hello FoodMan,

In general it is recommended to keep bounding box reasonably tight. Small difference in size will not affect performance significantly.

But for outdoor scenes distant objects can be included in the bounding box, which can increase its size many times compared to the size of the object of interest. This can degrade performance significantly.

[FoodMan]

ok I see... thanks Dmitry

f/

[glennn]

Hi Dmitry
I always set the bounding box to a tight close fit to the final mesh I need. 
Does deleting the point cloud data in the distance effect anything?  I sometimes just delete a lot of stray points outside the bounding box.
Cheers

[Alexey Pasumansky]

Hello glennn,

Editing point cloud will effect the result of geometry reconstruction in two cases:
1) Point cloud geometry type is selected in Build geometry dialog,
2) After removing some points optimization process is performed.

[matteoponzio]

Hey guys,
    I'm experiencing the same problem, when Photoscan decimates the mesh it stops processing.

I just know that I didn't have this problem before upgrading to Photoscan 0.8.5. (and I've a 4 GB RAM laptop    and that this problem only occurs when I want Target quality: High (no problems with Medium).

I'm very interested in getting input about how to get across this problem. I need to produce high quality models.

[FoodMan]

Matteoponzio...

4GB is very little these days...

f/

[Alexey Pasumansky]

Hello matteoponzio,

Could you please specify the number of images, resolution, object/scene type and reconstruction method?
And also please make sure you are using 64-bit version, just in case.

[matteoponzio]

Hi Alexey,
    I'm sorry for the big delay in replying.

An example of crashed reconstruction is
nr of photos: 63
resolution: 3872x2592
scene type: it is a 5-6 m long and up to 1,7 m high archaeological excavation trench
reconstruction method:
- Align photos: accuracy high, pair selection disabled;
build geometry: arbitrary, high, sharp, 500000, 0.5, 0.1;

It crashes when it decimates the model at the end of the reconstruction process.

My computer could handle same parametres, same photo resolution and even more pictures in 5-6 hours with the previous version of Photoscan.

What do you think it is?

we in archaeology love the possibility to produce models without need of monster-cameras, monster-softwares and monster-computers ... as several of the people in this thread    Photoscan should be able to work on normal computers too. Don't you think?

[Wishgranter]

Matteoponzio. the bigges problem for what is photogrametry just now accesible to wide audience is enought CPU power and amount of RAM. they will be problematic for some time, because of prices of RAM. not so CPUs.... if want and need help with recontruction for good prices ( must not buy own "super computer " ) send me your email to muzeumhb@hmail.com

[Alexey Pasumansky]

Hello matteoponzio,

Thank you for the information.

The memory consumption could be higher if the bounding box has large dimensions. Could you please check it is set correctly and doesn't cover unwanted areas.

If it is possible please also provide the log file so we could check the size of bounding box and the number of faces in the model.

Since you mentioned the memory error appears on the decimate mesh step you can try to use face count = 0 value to skip this step. However, there could be problems with the visualization of the millions of faces on the laptop graphic card.

[matteoponzio]

hi Alexey, sorry for the long delay in replying.

Here it's a new try with the last version of Photoscan.
The object is a medieval boat from a moat.

38 pictures
4288x2848 pixel
the object is 5 x 1,5 m
Align photos: accuracy high, pair selection disabled
build geometry: arbitrary, high, sharp, 0, 0.5, 0.1
the bounding box is set close to objects limits

I've tried with 1000000 faces and undecimated model (as you suggested).

not enough memory / Error: bad allocation    is the message for both tries.

I guess that the solution is to buy more RAM, isn't it?

Still, a friend of mine can make small models of this kind with a computer that is as weak as mine (4GB RAM) ... that's not fair!

I attach the log file, just in case.

[Wishgranter]

Hello.

If your friend can create models is on same level of DETAIL - HIGH ?

How much memory you have from my knowlege for high you will need 16 GB RAM.... if need process it send me PM to muzeumhb@gmail.com can help with that....