Messages

1

General / Re: Info on accuracy/resolution provided by Agisoft photoscan.

« on: May 30, 2017, 09:05:09 PM »

Can't you measure this your self? If you make a model from something with a known shape you can calculate the difference between the ground truth and the model to find the accuracy?

2

Feature Requests / Re: Command for statistical best selection of intrinsic parameters

« on: May 22, 2017, 08:04:05 PM »

I second this.

3

General / Re: How to capture photos for a piece of kelp?

« on: May 22, 2017, 08:00:35 PM »

Longer exposure times, overexpose the kelp by one stop or so to bring out more shadow detail.

Move the camera around the kelp.

Focus again for every shot. Autofocus works fine.

Shoot at f8-12, iso100-200

Print out some coded targets and place them around the object

4

General / Re: Pixelpeeper: a free utility to visually estimate image sharpness

« on: May 22, 2017, 07:45:20 PM »

Perhaps you could build in a functionality where we can opt in to have tagged images as well as some sharp ones from the same set uploaded to you. After you get a few thousand real world blurry images you could train a neural network to identify them and then make auto detection a part of the program? 

5

General / Re: Lack of point cloud data in well aligned areas.

« on: January 30, 2017, 05:35:24 PM »

This is just my anecdotal evidence, but I've found that underexposing overexsposed images by a few stops and increasing the contrast can do wonders for data sets that otherwiser have problems aligning. So that could at least be worth a try perhaps?

6

General / Re: Suggestions to improve model quality needed.

« on: January 30, 2017, 04:34:11 PM »

I know this might be a bit of a strech, but would you be willing to share your data set? I'd like to try and replicate the results from Alexey to see if my workflow will get the same results. That way I could know whether or not my workflow needs improvement.

7

General / Re: Noisy data

« on: January 14, 2015, 10:10:45 PM »

I have som (very limited) data that seems to suggest that lower ISO isn't always the best with photogrammetry. In my case with my DSLR I seem to get more points when I use iso 400 compared to 200.

As for your problem, you may want to try to process a copy of your raw data files and bump the contrast as far as it will go. This will hopefully give you a little more contrast in the areas you struggle with now and help photoscan detect more points. Simply change the pictures to the unprocessed ones when generating the texture.

Also make sure the plaster wall isn't overexposed, but I assume it's not.

8

General / Re: Using a GoPro & Photoscan

« on: January 14, 2015, 10:04:40 PM »

Just a tip: You might find a lens profile for the gopro for lightroom. You can then back process the images there and give photoscan the corrected images.

9

General / Re: Dual cameras for better results in Photoscan

« on: December 16, 2014, 02:49:20 AM »

There is a difference between subject matter:

If you are doing human scans then multiple cameras will be better (faster acquisition so less movement between photos).

If you do fully static objects then a single camera will give you exactly the same quality as using multiple cameras.

Capturing with two cameras at a known distance from each other has no benefit in Photoscan (apart from capturing twice as fast). This distance information is not used so you do not gain processing speed or accuracy.

Well consider this: If you take 40 shots of a model "free handed", chances are that some images might not align properly or that the angle or distance makes for various numbers of points successfully generated between image pairs. Now if you have two cameras positioned in an ideal angle, the success rate for these image pairs will always be 100%, though between n image pairs there might still be variance in terms of the number of successful points. I'll hypothesize that on average you'll get more points generated with a dual camera rig than a single camera with equal numbers of pictures in the data set. And I'll hypothesize that this will be true no matter if the subject is static or moving slightly. Do anyone know the ideal angle between two cameras BTW?

Capturing with two cameras at a known distance from each other has no benefit in Photoscan (apart from capturing twice as fast). This distance information is not used so you do not gain processing speed or accuracy.

In PhotoScan Pro version 1.1.0 it is possible to create scale-bars between pairs of camera centers and use this information for scaling and optimization purposes. However, the stereobase should not be negligibly small compared to the distance to object.

Do you know if there is any good documentation on this? I want to try it out, but I'm not sure I completely understand how it's supposed to work. Also, do you know the ideal positioning of the two cameras?

10

General / Re: Visible triangles in models

« on: December 14, 2014, 05:20:31 PM »

Surface type arbitrary, medium polygon count, source data is a dense cloud.

11

General / Visible triangles in models

« on: December 14, 2014, 04:46:27 AM »

Hi.
Recently the meshes I generate in photoscan have come out looking very "triangulated". As inm if you zoom in on the model, you can see a lot of very noticeable triangles. Examples:


https://skfb.ly/C6DU

I didn't have this problem when I used an older version of photoscan. Example:

https://skfb.ly/zIvT

The only difference between those two models are the photoscan versions. Is there a setting or something that cause this? Any insight would be appreciated.

Regards

12

General / Dual cameras for better results in Photoscan

« on: December 14, 2014, 01:38:48 AM »

So I came over this interesting blog post today:

http://blog.kiteandlightning.la/breakthrough-a-3d-head-scan-using-one-camera/

Here the author claims that using "stereo pairs when using photogrammetry results in superbly clean meshes".

Have anyone tried something like this?