How many cameras can I use on Photoscan?

[senzasquadre]

Hi guys,

I'm a student of Architecture and for my final thesis I need to do some models in Photoscan. In the specific case, I'm working on a cloister: I need to take several pictures of this, something like 2000, and I was quite disperate because I think it's impossible to work with all these pictures at the same time.

So, I tried to divide this model in four little models: I did the first and just to finish the dense cloud my laptop has to work almost 28 hours. I had 306 cameras, 18 mp each one, and I used this configuration:

Align photos:
-High
-Reference
-key point: 40000
-tie point: 4000

Dense Cloud
-Higher
-Aggressive

I have an HP with 12 GB RAM,
 Intel core i7 - 7500 u,
hdd 1tb,
amd r7 m440 graphics 2 gb and without,
windows 10

Is a pc problem or is another problem?
How many cameras can I use on Photoscan at the same tame? Are 300-400 cameras too much?

Plus, if I will be able to finish my models, how can I merge all these models in just one model?
Reading on the internet, I have understood that I can work with four different chunks in the same file and then I can merge them. Is correct?

Thank you so much and sorry for my english, I hope it's understandable
Greetings from Italy!

[SAV]

Hi senzasquadre,

Processing of pictures (in a single chunk) is limited by your hardware. Have a look at Photoscan's memory requirements.
http://www.agisoft.com/pdf/tips_and_tricks/PhotoScan_Memory_Requirements.pdf

Basically your computer's RAM will limit the number of pictures you can process in a single chunk. If you have large datasets, you will have to split them up in several smaller chunks and merge them at the end. Alternatively, you can run it on a more powerful machine that has more memory. No splitting needed then.

Note that for the first step (image alignment) PhotoScan is mainly using your CPU (=processor) and since the latest version also your GPU (= graphics card; not sure to which percentage though). If align photos is slow, you need to get a faster and/or CPU with more cores. Your i7 7500u is only a dual core processor. I suggest to get at least a i7 quadcore processor.
Here is a list of Intel's 7th Generation i7 CPUs for laptops:
https://ark.intel.com/products/series/95544/7th-Generation-Intel-Core-i7-Processors#@mobile

As you can see, your CPU is at the lower end of the list. The i7 7920HQ would be the most powerful one with 4 cores and higher clock speed.

For the computation of the dense point cloud, Photoscan uses your GPU. Your AMD R7 m440 graphics card is definitely a bottleneck for the dense point cloud quality (HIGHER) that you have chosen. You can reduce the quality to MEDIUM or split your project into several smaller chunks which you then process individually and merge in the end or get some better hardware. Nvidia GTX 1070 or GTX 1080 and GTX 1080 Ti are probably a good GPU choice.

Merging of model can be done in PhotoScan.
1. Workflow > Align chunks
2. Workflow > Merge chunks

Yes, you would have one project with several chunks that are then merged.

Btw, your English is VERY understandable 

Ciao
SAV

[Yoann Courtois]

Hi senzasquadre !

Here are some tips coming from my experience:
- PhotoScan is able to work with such amount of pictures (I've handled more than 3000 pictures at the same times), so no problem for your project, you may just have to take care about some things.
- To my mind, it's really really really better to align your whole set of pictures in the same chunk ! Indeed, you will be able to adjust more accurately your whole model as it forms a single bloc. Keep your alignment parameters, I'm sure you computer is OK to handle that. It may just take a while (a night or two of calculation) but you will clearly take advantage of it. By align in different small chunks you may have some distortions between them which will create problems at the edge when you will align chunks.
- After adjustment (on GCPs if you have), duplicate your chunk to densify by sectors using different bounding boxes (I would say between 4 and 8 chunks to have 300-500 pictures per area). Don't touch the pictures in any chunk, just change the bounding box from a chunk to the next one to from a chessboard.
- Indeed, your computer won't be able to densify everything at the same time. And even it was able to, it would take sooo long time.
- So then use a batch to densify your 4-8 chunks one after the other.
- For your densification parameters, as SAV said, I guess you may revise them downwards. "Higher" use the initial resolution of your pictures which is too high for the densification. Use "High" or even "Medium" will give you a nice result.

Question about that: What is your average ground sampling distance (GSD) ?

To summarize, make an effort for alignment (The whole set of pictures at the same time and "High" parameter) and keep calm for densification (by dividing the project into small boxes and use "High" or even "Medium" parameter).

Regards

[SAV]

Hi again senzasquadre,

It is a good idea to follow Yoann's advice and try to align all your images as a single chunk before splitting them up to build dense point clouds. Additionally, I suggest to use the SPLIT IN CHUNKS Python script to do so. Will make your life much easier. Here is where you can download it from:
http://wiki.agisoft.com/wiki/Python

Regards
SAV

[senzasquadre]

Thank you so much!!!  I will try to do these things and I hope to fix my problem - in the meanwhile, I tried to create the point cloud for my second model (using high, no more higher) and at the moment I have to wait 33:20:37 hours. Amazing

I have just some questions:

- To my mind, it's really really really better to align your whole set of pictures in the same chunk ! Indeed, you will be able to adjust more accurately your whole model as it forms a single bloc. Keep your alignment parameters, I'm sure you computer is OK to handle that. It may just take a while (a night or two of calculation) but you will clearly take advantage of it. By align in different small chunks you may have some distortions between them which will create problems at the edge when you will align chunks.
- After adjustment (on GCPs if you have), duplicate your chunk to densify by sectors using different bounding boxes (I would say between 4 and 8 chunks to have 300-500 pictures per area). Don't touch the pictures in any chunk, just change the bounding box from a chunk to the next one to from a chessboard.
- Indeed, your computer won't be able to densify everything at the same time. And even it was able to, it would take sooo long time.
- So then use a batch to densify your 4-8 chunks one after the other.

I have a total of 2217 pictures and to align 412 pictures my laptop took 6 hours, less and more. Do you think is still better to start again with a new chunk, at this point? More important, in your opinion is my pc capable to align these huge numbers of images?

Additionally, I suggest to use the SPLIT IN CHUNKS Python script to do so. Will make your life much easier. Here is where you can download it from:
http://wiki.agisoft.com/wiki/Python

Regards
SAV

I'm happy to learn everything will make my life much easier, but I'm not sure to know what exactly a script is and how can I use it in Photoscan. Is there a tutorial somewhere in this site? I don't have Python installed on my pc, is a problem or is a script that I can use inside of Photoscan - in this case, where can I put it?

Thank you so much again!

[SAV]

Hi senzasquadre,

Photo alignment mainly depends on the number of photos and practically does not depend on the resolution of individual photos. According to PhotoScan's memory requirement PDF, you'll need at least 10GB of RAM to align 2000 photos. Therefore, it should work on your laptop with 12 GB RAM. It will take quite some time though because of your dual core processor.

Regarding the Python Script. Simply download it from the Wiki and save it as PS130_split_in_chunks_dialog.py file on your hard drive.

In PhotoScan, go to your console pane and click on the Run Python script icon (third from left, looks like a white page with blue < > symbols on top). If you cannot see the console pane, go to VIEW > PANES and make sure that Console is activated.
After you clicked on the Run Python Script, choose the folder where you just stored the PS130_split_in_chunks_dialog.py file and then click OK. Nothing will happen, but you should see a new Custom menu bar appearing at the very top which has one point called Split in chunks. Click on it to run the script.

Hope that works for you.

Regards
SAV

[senzasquadre]

Regarding the Python Script. Simply download it from the Wiki and save it as PS130_split_in_chunks_dialog.py file on your hard drive.

Thank you again. I tried to download this, but I discovered I have a Photoscan 1.2.0 and the Python Script 1.3.0 doesn't work on my pc. I found another Python Script on the internet, but it's for Photoscan 1.1.0. For some reason, do you know where - if it still exists - is possible to find the script for 1.2.0 ?

Otherwise, I will try to follow Yoann Courtois advices, duplicating the chunks to densify them in separated sectors

[Yoann Courtois]

You should update PhotoScan on your computer to 1.3 firstly !
Indeed, no need to pay something more, and you will enjoy new optimizations of the software and so really quicker processing time !

As well, using Split_in_Chunks script will help you to easily duplicate your project and densify your point cloud quicker and without a lot of RAM.

Aligning a set of 2000 pictures should take a lot of time, and even more with a dual core processor on PS 1.2. By updating the software, the alignment step will take really less time and it will also use your GPU during the calculation.
Anyway, aligning your whole set of pictures at the same time is really needed if you aim an accurate reconstruction.

Regards