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 ... 12 13 [14]
Bug Reports / Re: 0.9.1 Build 1647 crashes graphics
« on: March 29, 2013, 09:37:57 AM »
Hello David,

Thank you for reporting.

Could you please check that 0.9.0 is working fine under the same system configuration.

I removed 0.9.1, loaded 0.9.0 and verified that 0.9.0 is working fine under the same configuration.

I removed 0.9.0 and loaded 0.9.1 Build 1647 and the same problem occurs.

I previously updated the graphics driver to 12/19/2012. This did not affect the problem.

Bug Reports / 0.9.1 Build 1647 crashes graphics
« on: March 28, 2013, 10:41:21 PM »
(Also reported in Agisoft PhotoScan 0.9.1 pre-release thread.)

Build Geometry in 1647 Standard Version crashes the AMD 7770 graphics on my system when OpenCL is used. I have not had the problem with previous builds. Align Photos is fine. A second or so after Build Geometry is started the screen goes black and then a message appears saying graphics was shut down and restarted. PhotoScan continues to run but with low CPU usage and no GPU usage. Not obvious any progress is made. The behaviour is repeatable.

When OpenCL is disabled Build Geometry runs okay.

Machine is running Windows 7, Intel i7-3770, 16 GB memory, AMD Radeon HD 7700 graphics with 3 GB memory, AMD driver 9/27/2012. I checked for driver updates from AMD and driver is shown as most recent.

I normally run PhotoScan with OpenCL enabled and 6/8 Active Cores.

Correction: Graphics has 2 GB memory, not 3 GB.

"Depth of field" does not mean everything within the depth of field is in focus and everything outside the depth of field is equally out of focus.

Keep in mind that there is only one plane (actually usually a curved surface) at which objects are in focus and therefore maximum sharpness for the lens. The distance from the lens to the plane/surface is the distance the lens is focused at. Away from that plane/surface objects will be blurred, with the amount of out of focus blurring increasing the further the object is from the plane/surface.

The overall sharpness of an object away from the plane/surface depends on two factors. One is the inherent sharpness of the lens for which there is an optimum f-stop. A higher numerical f-stop will reduce this part of sharpness. The other factor is the amount of blur due to the object being out of focus. This amount of blur increases the further the object is from the "focus distance" but decreases with higher numerical f-stop. So if the object being photographed is sufficently close to planar then for maximum sharpness use the optimum f-stop for the lens (assuming there isn't blurring due to motion, etc). But if object isn't close to planar than a higher f-stop will result in improved sharpness for the portions of the object away from the focus distance while decreasing sharpness at the focus distance. But there may be a higher f-stop at which the decrease in lens sharpness is greater than the decrease in out-of-focus blurring. The more non-planar the object is the higher the f-stop to use.

General / Re: before buying some tip for best quality
« on: February 23, 2013, 12:05:31 AM »
You can use LightRoom to adjust the photos so the shadow details show up. Either use the Whites/Highlights/Shadows/Blacks sliders or adjust the tone curve.

General / Re: before buying some tip for best quality
« on: February 21, 2013, 12:03:40 AM »
My experience is better results are obtained with PhotoScan if the lens distortion correction in LightRoom, OpticsPro or similar software is not used.

I have experimented with using chromatic aberration correction. Depending on the lens there can be an improvement in PhotoScan results.

Vignetting correction does not seem to have much effect. I have used adjustments during "development" of the RAW photos to improve the amount of detail in shadows and highlights, and this has improved PhotoScan results in some cases. It is good to have reasonable consistency across images.

I use a Canon DSLR with a 18-55 mm zoom lens and a 40 mm lens. The zoom lens is most commonly used at 18 mm (about 28 mm full frame equivalent) and has also been used at 24 mm. The choice depends on the shape of the object being photographed and the amount of space around the object.

I shoot RAW and convert to TIFF with the DSLR. I previously used a camera which only produces JPEG, and I used the highest resolution and quality JPEG available.

Bug Reports / Chunk Align, Camera Method - Batch Only
« on: October 30, 2012, 06:52:08 AM »
The only way I can use the Camera Method for aligning chunks is to go through Batch Process. Then I can select it. If I select Align Chunks directly from the Workflow menu then I don't have a choice of methods to use. Only the Points Method is available. The Camera Method is several orders of magnitude faster for my current project and appears to be as accurate given the large number of cameras common between chunks.

Bug Reports / Undo and Redo not working
« on: October 29, 2012, 04:43:50 AM »
The Undo and Redo icons are never active for me. I'm using PhotoScan Standard Version 0.9.0 build 1586 (64 bit) with Windows 7.

Feature Requests / Start button for batch processing
« on: October 29, 2012, 04:40:29 AM »
Request - A separate "Start" button to begin executing batch processing instead of having it start when the Batch Process window is closed with the OK button.

Rationale - Ability to close the Batch Process window without starting processing. I frequently need to close the Batch Process window while adding tasks to either check something using another window. Currently I have to click OK, have batch processing start, and then cancel batch processing.

General / Re: Question about resolution
« on: September 27, 2012, 04:52:35 PM »
More faces and vertices are needed.

Did you Decimate the model? If so rebuild the model and increase the "Target face count" in "Decimate". Try 1 million faces.

If you did not Decimate the model then rebuild the model and increase "Face count" in "Build Geometry".

Feature Requests / Re: 6 DOF controller compatibility
« on: September 27, 2012, 12:08:04 AM »
I support this request. I use  3DConnexion Space Navigator with other software. It would be more efficient to be able to do translations and rotations of the display without having to switch from editing.

General / Re: Rendering Point Cloud
« on: September 15, 2012, 11:09:04 PM »
Method I've used to obtain the vertex points from a mesh created by PhotoScan.
1) Export the mesh from PhotoScan as a .obj file.
2) Import the .obj file into Rhino.
3) Extract the vertices of the mesh as points us the Rhino command ExtractPt. This can take a while.
4) Export the vertices in the desired format if they are to be used in other software.
I've used this method for meshes with several million vertices.

General / OpenCL Prefrence Settings: Intel i7-3770, AMD Radeon HD 7770 2GB
« on: September 15, 2012, 07:28:38 AM »
I'm a new user of PhotoScan and I experimented to find the "best" settings for OpenCL with my hardware. Bottom line is performance is very close when OpenCL is turned on and number of active CPU cores is set to 4, 5, 6 or 7 for my 8 thread CPU. Execution time for "Build Geometry" is cut in half and "Total Performance" more than doubles. Now for some details.

PhotoScan 0.8.5 build 1423 (64 bit)

HP Pavillion H9
Motherboard: Petragon 2AD5, Intel Z75 chipset
CPU: Intel i7-3770 CPU, rated 3.4GHz (3.9GHz max observed), 4 cores / 8 threads
Graphics: AMD Radeon HD 3770, 2GB memory (HP version), 1000MHz, Capeverde 10 core processor
Memory: 12GB

I used the first 10 photos in the sample01 download set. Align Photos was run with Accuracy: High.
The bounding box was reduced to be tight to the statue and plinth. No masking was used.

The OpenCL settings box shows a maximum of 8 active CPU cores.

Build Geometry (Arbitrary, High) was run 28 times with various combinations of OpenCL off/on and number of active CPU cores.

Execution Time (seconds) and  Performance (million samples/sec) were recorded for Device 1 (CPU), Device 2 (Capeverde, graphics processor) and Total Performance. There was considerable variation in the total times for repeat runs with the same settings, up to 12%. Performance results were much tighter. Times given below are typical.

OpenCL,  Active CPU Cores, Time (sec), CPU (ms/s), GPU (ms/s), Total (ms/s)
Off, Any, 315, 130, 0, 130
On, 8, 219, 127, 80, 205
On, 7, 155, 127, 175, 302
On, 6, 158, 126, 187, 312
On, 5, 155, 124, 190, 315
On, 4, 150, 124, 192, 317
On, 2, 172, 108, 195, 303
On, 1, 214, 73, 200, 272
On, 0, 237, 0, 197, 197

Clearly turning OpenCL on and a setting for number of active CPU cores of 4 to 7 yields the best performance, with 4 slightly better than 5, 6, or 7.

The percent of utilization for the CPU cores was observed using Open Hardware Monitor and Windows Performance Monitor. The work was distributed across the cores regardless of the number of active CPU cores setting selected. This was true both with OpenCL on and off.

Pages: 1 ... 12 13 [14]