Zhongyi (Mitakon) Super Macro Lens (1 - 5x)
Moderators: rjlittlefield, ChrisR, Chris S., Pau
-
- Posts: 3438
- Joined: Sat Nov 20, 2010 10:40 am
- Location: Santa Clara, CA, USA
- Contact:
I think the problem may be that you are not actually stitching with planar motion. When I select planar motion, the program eliminates any pitch and yaw adjustments since of course there can be none if the camera is operating in planar motion! Roll is still available since that is just a simple rotation of the final image.
why do you think that way? please see my last posts as why that, "so if the software is making some tiles larger than others, the output is being distorted", is not the case.ray_parkhurst wrote:Not really. Actually his comments reinforce mine. Each of the tiles is shot at the same magnification, so if the software is making some tiles larger than others, the output is being distorted. This should not happen independent of how flat the subject is.mjkzz wrote:I see Rik has comment and I think it will answer yours to avoid duplicate thoughts
-
- Posts: 3438
- Joined: Sat Nov 20, 2010 10:40 am
- Location: Santa Clara, CA, USA
- Contact:
I am assuming the stretching of the image is due to the tiles in that region being scaled larger vs the other tiles, and in that case those tiles are distorted vs the unscaled ones.mjkzz wrote:why do you think that way? please see my last posts as why that, "so if the software is making some tiles larger than others, the output is being distorted", is not the case.ray_parkhurst wrote:Not really. Actually his comments reinforce mine. Each of the tiles is shot at the same magnification, so if the software is making some tiles larger than others, the output is being distorted. This should not happen independent of how flat the subject is.mjkzz wrote:
I see Rik has comment and I think it will answer yours to avoid duplicate thoughts
I use auto detect, which means, ICE will make a decision. If you can always stitch something by forcing ICE to use planar motion, then I think you should try something more 3D like.ray_parkhurst wrote:I think the problem may be that you are not actually stitching with planar motion. When I select planar motion, the program eliminates any pitch and yaw adjustments since of course there can be none if the camera is operating in planar motion! Roll is still available since that is just a simple rotation of the final image.
No, the "stretching" part is over simplification of what actually is being done. The fact that ICE allow user to interact with final result with pose parameters means it is due to projecting a 3D model onto 2D surface.ray_parkhurst wrote:I am assuming the stretching of the image is due to the tiles in that region being scaled larger vs the other tiles, and in that case those tiles are distorted vs the unscaled ones.mjkzz wrote:why do you think that way? please see my last posts as why that, "so if the software is making some tiles larger than others, the output is being distorted", is not the case.ray_parkhurst wrote:Not really. Actually his comments reinforce mine. Each of the tiles is shot at the same magnification, so if the software is making some tiles larger than others, the output is being distorted. This should not happen independent of how flat the subject is.mjkzz wrote:
I see Rik has comment and I think it will answer yours to avoid duplicate thoughts
-
- Posts: 3438
- Joined: Sat Nov 20, 2010 10:40 am
- Location: Santa Clara, CA, USA
- Contact:
The 3D nature of the subject has nothing to do with the choice of stitching method. The selection of camera motion is indeed what it says, camera motion, not subject characteristics. Since your camera is moving in XY on rails, it is moving in a plane, so the appropriate selection is planar motion. If you let the program select something else, it is an error.mjkzz wrote:I use auto detect, which means, ICE will make a decision. If you can always stitch something by forcing ICE to use planar motion, then I think you should try something more 3D like.ray_parkhurst wrote:I think the problem may be that you are not actually stitching with planar motion. When I select planar motion, the program eliminates any pitch and yaw adjustments since of course there can be none if the camera is operating in planar motion! Roll is still available since that is just a simple rotation of the final image.
No, it is not an error, it is a choice.ray_parkhurst wrote:The 3D nature of the subject has nothing to do with the choice of stitching method. The selection of camera motion is indeed what it says, camera motion, not subject characteristics. Since your camera is moving in XY on rails, it is moving in a plane, so the appropriate selection is planar motion. If you let the program select something else, it is an error.mjkzz wrote:I use auto detect, which means, ICE will make a decision. If you can always stitch something by forcing ICE to use planar motion, then I think you should try something more 3D like.ray_parkhurst wrote:I think the problem may be that you are not actually stitching with planar motion. When I select planar motion, the program eliminates any pitch and yaw adjustments since of course there can be none if the camera is operating in planar motion! Roll is still available since that is just a simple rotation of the final image.
Having ICE to decide means I want to extract more info from ICE, how it works. If after computation, ICE decides everything seems to be planar motion, then we have a perfect stitch. If ICE decides, say perspective, it means we have some issue and we need to do something about it, say, increase overlap. If we are not happy with final results, say perspective one, we can reject it.
See having it set to auto, we can get ICE to report more about the situation. It is not an error, it is a choice.
Further more, I do not think it is a choice of stitching method, it is an restriction , if after computation, the final result does not suggest planar motion, ICE will just say it can not stitch.The 3D nature of the subject has nothing to do with the choice of stitching method. The selection of camera motion is indeed what it says, camera motion, not subject characteristics. Since your camera is moving in XY on rails, it is moving in a plane, so the appropriate selection is planar motion. If you let the program select something else, it is an error.
Last edited by mjkzz on Mon May 27, 2019 5:38 pm, edited 1 time in total.
Ray, I think I see where we have different opinions. You think the camera motion selection as method, I see it as restriction.
Either way, data presented to ICE is the same, after computation, if your selection is planar motion and computation does not support that, ICE will just say no, can not stitch. If your selection is perspective, ICE will say, oh ok, here you are, but it will be up to you to take or reject it.
So to me, that is a restriction because regardless selection, computation is the same.
Either way, data presented to ICE is the same, after computation, if your selection is planar motion and computation does not support that, ICE will just say no, can not stitch. If your selection is perspective, ICE will say, oh ok, here you are, but it will be up to you to take or reject it.
So to me, that is a restriction because regardless selection, computation is the same.
-
- Posts: 3438
- Joined: Sat Nov 20, 2010 10:40 am
- Location: Santa Clara, CA, USA
- Contact:
Certainly it is restricting the manipulations that the software can do to the image, but the restrictions are appropriate. I shudder when I hear that you are allowing ICE to decide how to stitch when indeed you know the camera motion was planar.mjkzz wrote:Ray, I think I see where we have different opinions. You think the camera motion selection as method, I see it as restriction.
Either way, data presented to ICE is the same, after computation, if your selection is planar motion and computation does not support that, ICE will just say no, can not stitch. If your selection is perspective, ICE will say, oh ok, here you are, but it will be up to you to take or reject it.
So to me, that is a restriction because regardless selection, computation is the same.
edited to add: for the beetle, you were only successful in stitching with planar motion. Doesn't that tell you something about the correct method to use?
ICE kept deciding there is a yaw rotation, though small, less than half degree, it still is.ray_parkhurst wrote:Certainly it is restricting the manipulations that the software can do to the image, but the restrictions are appropriate. I shudder when I hear that you are allowing ICE to decide how to stitch when indeed you know the camera motion was planar.mjkzz wrote:Ray, I think I see where we have different opinions. You think the camera motion selection as method, I see it as restriction.
Either way, data presented to ICE is the same, after computation, if your selection is planar motion and computation does not support that, ICE will just say no, can not stitch. If your selection is perspective, ICE will say, oh ok, here you are, but it will be up to you to take or reject it.
So to me, that is a restriction because regardless selection, computation is the same.
edited to add: for the beetle, you were only successful in stitching with planar motion. Doesn't that tell you something about the correct method to use?
I do know the algorithms estimating camera position and pose are very robust, even back in 2008. So letting ICE to decide is good way to check "systematic" error. Maybe the XY table is tilted (as later 10x stitching shows, I need to increase about 60um for some section of Velvet ant), so it is NOT true planar motion according to computation, even though physically it "seems" so.
I see your point, by pre-selecting a choice will make/force ICE to compute in certain ways (for example, forcing all camera poses to be the same, maybe some averaged pose), but we are going to have distorted image if our system has some flaws.
Here is a video, the first part shows Agisoft's Metashape and the next shows Zephyr Free. The cameras are kind of faint blue, so maybe zoom it up to full screen. Obviously, the Zephyr one is better as the final 3D model is "complete" (limited by one directional shoot)
I think Agisoft Metashape will do just as good once images shot from different angles are present
https://youtu.be/DlYOwcHbKCU
I think Agisoft Metashape will do just as good once images shot from different angles are present
https://youtu.be/DlYOwcHbKCU