[ContextCapture] - General guidelines on usage

Hi,

We have been starting to test CC to see if it can be useful for us in our business. On this a couple of questions have came up:

1. Photogroups

Are there any advantages/disadvantages to have several photogroups, even if all pics have same properties? Let's say I have 1000 pics - will there be any difference (processing speed, quality) if I would split these pics into several photogroups?

2. Tiling

Sort of same question as above, but on tiling: Any difference on quality and/or processing time depending on selection of (level of) tiling? Should we try to have as few tiles as possible, or could it possibly be the opposite => that MANY tiles are preferable?

3. Tiling algorithm

We'be been a bit puzzled on what it is that actually "triggers" that tiling is necessary. Let me explain: Some models we've done have been from quite large amount of (hires) pics, resulting in NO need for tiling. Other models have been fewer pics, but still resulted in need of tiling. How come..?

4. Tie points

It is interesting to note/see the resulting number of calculated tie points after aerotriangulation. It can differ severly from (in principle) same set of pics (for example: Before/after balancing pic histograms). It can differ a factor two. But strangely enough, more tie points doesn't necessarily make a better model... So the question: How should one "interpret" the resulting no:s of tie points? Can any "conclusions" be made from the calculated number?

5. User tie points

It is possible to create user tie points. But why...? I mean, in what situation is that useful/necessary? In principle: Is there any reason why I should put the time to add some manual tie point to the [many thousands] that are created in the aerotriangulation? Could A FEW manual tie points make any major difference to the quality of the created model?

Thanks a lot in advance for answers/comments!  :-)

Parents
  • [Bump] - Is there possibly someone who has some ideas around my questions above...? Felix - you seems to be skilled in this tool! :-) Any comments?
  • Hi,

    1. Photogroups

    Are there any advantages/disadvantages to have several photogroups, even if all pics have same properties? Let's say I have 1000 pics - will there be any difference (processing speed, quality) if I would split these pics into several photogroups?

    Answer

    ContextCapture will process the photos and automatically determine the photogroups based on the camera properties (image dimensions, sensor size, focal length etc.) Splitting the subdividing the photogroups any further will not improve the performance /quality. If images share the same optical properties it is recommended that they be left in a single photgroup.

    2. Tiling

    Sort of same question as above, but on tiling: Any difference on quality and/or processing time depending on selection of (level of) tiling? Should we try to have as few tiles as possible, or could it possibly be the opposite => that MANY tiles are preferable?

    Answer

    Producing big tiles (and less numerous) tiles the processing will go slightly faster but you must be careful and take note of the expected RAM consumption

    3. Tiling algorithm

    We’ve been a bit puzzled on what it is that actually "triggers" that tiling is necessary. Let me explain: Some models we've done have been from quite large amount of (hires) pics, resulting in NO need for tiling. Other models have been fewer pics, but still resulted in need of tiling. How come..?

    Answer

    This is dependent on the number of the images but also their size and resolution

    4. Tie points

    It is interesting to note/see the resulting number of calculated tie points after aerotriangulation. It can differ severely from (in principle) same set of pics (for example: Before/after balancing pic histograms). It can differ a factor two. But strangely enough, more tie points doesn't necessarily make a better model... So the question: How should one "interpret" the resulting no:s of tie points? Can any "conclusions" be made from the calculated number?

    Answer

    Unless you notice a real lack of tie-point in the 3D-View, you shouldn’t be concerned about that as we go through a “denser” computation during reconstruction.
    You can also find some relevant explanations in the AT-report

     

    5. User tie points

    It is possible to create user tie points. But why...? I mean, in what situation is that useful/necessary? In principle: Is there any reason why I should put the time to add some manual tie point to the [many thousands] that are created in the aerotriangulation? Could A FEW manual tie points make any major difference to the quality of the created model?

    Answer

    There are cases where automatic tie points are required to stitch sub-parts of the dataset that would not be automatically aligned (huge scale difference, lack of overlap etc.). This is not needed in most cases. It can also be used to add positional or scale constraints

Reply
  • Hi,

    1. Photogroups

    Are there any advantages/disadvantages to have several photogroups, even if all pics have same properties? Let's say I have 1000 pics - will there be any difference (processing speed, quality) if I would split these pics into several photogroups?

    Answer

    ContextCapture will process the photos and automatically determine the photogroups based on the camera properties (image dimensions, sensor size, focal length etc.) Splitting the subdividing the photogroups any further will not improve the performance /quality. If images share the same optical properties it is recommended that they be left in a single photgroup.

    2. Tiling

    Sort of same question as above, but on tiling: Any difference on quality and/or processing time depending on selection of (level of) tiling? Should we try to have as few tiles as possible, or could it possibly be the opposite => that MANY tiles are preferable?

    Answer

    Producing big tiles (and less numerous) tiles the processing will go slightly faster but you must be careful and take note of the expected RAM consumption

    3. Tiling algorithm

    We’ve been a bit puzzled on what it is that actually "triggers" that tiling is necessary. Let me explain: Some models we've done have been from quite large amount of (hires) pics, resulting in NO need for tiling. Other models have been fewer pics, but still resulted in need of tiling. How come..?

    Answer

    This is dependent on the number of the images but also their size and resolution

    4. Tie points

    It is interesting to note/see the resulting number of calculated tie points after aerotriangulation. It can differ severely from (in principle) same set of pics (for example: Before/after balancing pic histograms). It can differ a factor two. But strangely enough, more tie points doesn't necessarily make a better model... So the question: How should one "interpret" the resulting no:s of tie points? Can any "conclusions" be made from the calculated number?

    Answer

    Unless you notice a real lack of tie-point in the 3D-View, you shouldn’t be concerned about that as we go through a “denser” computation during reconstruction.
    You can also find some relevant explanations in the AT-report

     

    5. User tie points

    It is possible to create user tie points. But why...? I mean, in what situation is that useful/necessary? In principle: Is there any reason why I should put the time to add some manual tie point to the [many thousands] that are created in the aerotriangulation? Could A FEW manual tie points make any major difference to the quality of the created model?

    Answer

    There are cases where automatic tie points are required to stitch sub-parts of the dataset that would not be automatically aligned (huge scale difference, lack of overlap etc.). This is not needed in most cases. It can also be used to add positional or scale constraints

Children