Show Posts

This section allows you to view all posts made by this member. Note that you can only see posts made in areas you currently have access to.

Messages - David Cockey

Pages: 1 ... 9 10 [11] 12 13 14
Feature Requests / Duplicate Photo Finder
« on: November 26, 2013, 03:40:07 PM »
A tool which finds duplicate photos/cameras in a chunk based on same name.

A chunk can have duplicate photos for several reasons including:
- same photos were in two different chunks which were merged
- photo was added multiple times to chunk by mistake

Also desirable but would be the ability to identify photos share the camera location and orientation within a specified tolerance. I sometimes take a second if I suspect a problem with the first, and then forget to delete the first photo.

General / Mesh Build in PhotoScan 1.0 - My Workflow
« on: November 26, 2013, 03:11:29 PM »
PhotoScan 1.0 (pre-release available in ) separates building the mesh from building the dense cloud. Below is a summary of the workflow I'm using in projects involving objects such as boats.

Mesh Creation in PhotoScan 1.0

1)  Trim significant noise and unwanted parts such as supports from the Dense Point Cloud before creating the mesh. The undo arrow in the icon toolbar at the top can be used to back up during the trimming process. If significant trimming is needed it may be best to create a copy of the chunk, and then trim the copied chunk.

2)  Click on Build Mesh in the Workflow and set the Build Mesh parameters:
Surface type:      Arbitrary
Source data:      Dense cloud
Polygon count:   40000000  (40 million, or a similar very large number)
Interpolation:   Enabled (default)
Click on okay and build the mesh.

3)  The mesh can be decimated in a separate operation if a smaller mesh is needed than the result from Build Mesh. The Decimate Mesh tool is found under the Tools menu.
      > Mesh
         > Decimate Mesh
Specify the number of faces desired and click okay.

4)  Build Texture if needed. A reasonable choice for the settings is:
   Mapping mode:   Generic
   Blending mode:   Mosaic (default)
   Texture size/count:   2048 x 1
      Enable color correction box not checked

Specifying a very large polygon count in Build Mesh results in the smallest amount of memory being used during the mesh generation process, contrary to what may seem obvious. The size of the mesh initially created during Build Mesh depends only on the number of points in the Dense Cloud and the geometry of those points, not on the value specified for Polygon count. The number of faces in the initial mesh is then compared to the specified Polygon count value, and the initial mesh is then decimated if needed so that the number of faces does not exceed the specified Polygon count value. Mesh decimation can use more memory than used for the creation of the initial mesh.

If the desired size of the decimated mesh is not certain, copy the chunk and then decimate the mesh in the copied chunk. If a different size mesh is desired then the original chunk can be copied again and the mesh in the second chunk decimated to the different size.

Enabled is generally the best choice for the mesh interpolation type for objects with strong three-dimensional shape.. Build Texture may not work properly if the mesh was created with Interpolation Disabled. Extrapolation attempts to create a closed mesh without any holes, and the result may not be the desired shape.

Texture should be created if the file is to be exported as a .obj and the mesh needs to have visual texture. Texture can also result more apparent detail and a more pleasing appearance with a relatively coarse mesh. If color correction is enabled for Build Texture the execution time may become much longer, particularly with a large number of photos.

David Cockey
26 November 2013

General / Re: Agisoft PhotoScan 1.0.0 pre-release
« on: November 25, 2013, 10:52:44 PM »
Using Interpolation solved the problem, though it didn't short the time to build the texture. The long time is probably due to the large number of photos.

Thanks for the help!

General / Re: What exactly do the reconstruction Quality labels mean?
« on: November 25, 2013, 10:24:55 PM »
My guess is the increments of 1, 2, 4, 8, 16 are inherent in the algorithms used.

General / Re: Agisoft PhotoScan 1.0.0 pre-release
« on: November 25, 2013, 10:01:25 PM »
Attached is what I see with the View Mesh UVs dialog box for the 2 million face mesh. Same thing with the other meshes.

I used Gradual Edit at 98% Connected Components with the 200,000 face mesh and removed the few small polygons. Currently rebuilding the texture which takes about 35 minutes.

Update: Result is all black texture.

General / Re: Agisoft PhotoScan 1.0.0 pre-release
« on: November 25, 2013, 09:21:34 PM »
I just tried decimating the mesh to 200,000 faces and building texture. The result is all black.

I have two similar chunks with 20 million and 2 million face meshes of each. The texture results for those also range from small "pixels" to large "pixels" to black.

General / Re: Agisoft PhotoScan 1.0.0 pre-release
« on: November 25, 2013, 08:04:08 PM »
Major Texturing Problem

Build Texture creates either black or very pixelated textures. I'm using Build 1774, Standard Edition, 64 bit, Windows 8.1.

Attached are the screen shots of the textured meshes for copies of the same chunk, 12,314,183 points in the Dense Point Cloud, meshes with 10 million and 2 million faces. Also attached is a screen capture of the 2 million face shaded mesh.

Texturing was done using:
Blending mode  Mosaic
Texture size      2048
Texture count    1
Color correction No

General / Re: Photo Realignment
« on: November 23, 2013, 05:21:16 PM »
I have reset the alignment and then realigned a single camera and the result is correct relative to the other images. So it appears that realigned images are realigned relative to the other images in the chunk.

General / Re: High demands need a high end computer. Help please
« on: November 12, 2013, 11:01:49 PM »
Increased memory only helps if you are running out of memory.

CPU speed is a major factor in how time align photos takes.

General / Re: High demands need a high end computer. Help please
« on: November 12, 2013, 02:18:19 AM »
GPU is not used during for alignment, only for creating the dense surface point cloud. Alignment uses only the CPU.

How much memory is currently being used during the alignment? If memory is available additional memory will not increase speed. If no memory is available then the system may be swapping data in and out of memory which drastically increases processing time. In Windows you can use the Performance tab in Task Manager to monitor memory and CPU usage. Performance Manager can be started for more detailed information. I use Open Hardware Monitor to monitor GPU usage as well as CPU and memory usage. It also shows temperatures if they are available.

One way to significantly reduce processing time:

1) Break the pipeline into chunks which overlap, and have several photos used in both chunks of each overlapping pair.

2) Align the photos in each chunk separately. Save after each chunk is aligned so that you won't lose everything if the computer crashes for some reason.

3) Align the chunks using the "camera" method. This is very quick but requires that each chunk have several photos which are used in at least one other chunk.

General / Re: Investing on a PC
« on: November 02, 2013, 09:45:20 PM »
but PS only loaded GPU in mesh generation.

My experience with 1.0 is the GPU is used during point cloud creation, not mesh generation.

General / Re: Software for reverse engineering?
« on: November 02, 2013, 09:43:15 PM »
I use standard Rhino commands and functions, no plug-ins, for going from point clouds and meshes to 3D models and lines. I have not seen anything which looks better for this type of work. Also, Rhino is very widely used for boat design so it serves a dual function for me.

I'm not familiar with PolyCAD so I can't comment on it.

General / Re: Software for reverse engineering?
« on: November 02, 2013, 06:50:33 PM »
I use Rhino to create 3D models and 2D drawings of historic boats starting with point clouds and meshes from PhotoScan. The method used depends on the shape of the hull and the completeness and amount of noise in the Photoscan point clouds and meshes.

Methods used to obtain NURBS curves in Rhino:

1) Cut a section through the PS mesh which results in a polyline. the polyline is rebuilt into a degree 3 NURBS using Rebuild or RebuildCrvNonUniform, and the resulting curve is smoothed and faired using FitCrv, Fair, Rebuild and/or RebuildCrvNonUniform.

2) Cut a section through the PS mesh which results in a polyline. Create a curve using InterCrv based on selected points along the polyline. Smooth and fair if needed.

Methods used to obtain NURBS surfaces in Rhino:

1) Create a network of NURBS curves using the methods described above, and then create the NURBS surface using one of the standard Rhino commands for going from curves to surfaces.

2) Create a network of NURBS curves using the methods described above, and then create the NURBS surface using the Patch command in Rhino. This can require experimentation, and I typically need to create a starting surface for Patch which has the approximate shape of the final surface.

3) Use the Patch command with all or part of the point cloud as an input.

My experience is that creating a good NURBS model from the mesh/point cloud of a boat hull usually requires more user time than required for obtaining the mesh/point cloud in PhotoScan. And that is for someone who is familiar with Rhino and modeling boat shapes in Rhino.

General / Re: Getting section cuts from the models
« on: November 02, 2013, 01:48:51 AM »
I use Rhino to create digital models of boats based on point clouds and meshes from PhotoScan. I import the mesh and/point cloud into a Rhino model and save it as a Rhino .3dm file. I do the Rhino work in a separate .3dm file. I use "Worksession" in rhino to "attach" the .3dm files with the needed meshes/point clouds so they can be used as references without adding to the size of the file with the modeling data.

Sufficient memory is needed so that Rhino does not start swapping data between memory and disk. How much memory depends on the number and size of the meshes/point clouds. I usually use meshes rather than point clouds.

General / Re: Workstation question
« on: November 01, 2013, 06:46:33 PM »
The GPU is only used for building the dense point cloud. CPU only is used for aligning, building mesh, building texture, etc.

I have a one year old desktop with an Intel i7 3770 CPU and an AMD Radeon HD 7770 graphics card. My experience is very similar to hengefjes. The GPU contributes significantly during building the dense point cloud. Cards similar to the card in my desktop currently sell in the US for slightly over $100.

Pages: 1 ... 9 10 [11] 12 13 14