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.

Topics - chadfx

Pages: [1]
Bug Reports / UI super slow when viewing detailed shaded masks
« on: May 06, 2014, 10:46:02 AM »
Ok, maybe not exactly a bug, but this seemed like the best place to ask about an issue.

In the MacOS version, if I import a set of masks that have a some noisy areas, the UI gets super slow when displaying those masks and trying to edit them further. I've been creating the masks through some keyer tools in Nuke, and they aren't always super clean extractions, resulting in visual noise in the form of lots of little white pixels scattered about it some areas.

Once I clean up the noise a bit, the UI gets much faster and responsive. So the basic idea of a lot of highly detailed outlines in the overlay is very bad news for interactivity.

Is there a way that Photoscan is drawing the overlay for the mask outlines in the UI that has some issues when trying to draw a LOT of them? I can push around millions of polygons without a problem, but a bunch of 2D outlines is a major slowdown.

It reminds me of an old issue we had on SGI machines, push a bit too much into the UI overlay graphics on the machine, and it would grind to a crawl...which otherwise was quite responsive for its time.

Anyway, just wondering if there was a way to fix that issue, because it really makes it hard to use the software in this case.

thanks! -Chad

General / Crowdsourcing Masks...
« on: April 21, 2014, 05:47:03 PM »
Well this is a novel idea...why make all of those tedious image masks all by yourself? Why not ask the public for help?

Bug Reports / PS 1.02 Import Mesh not parsing OBJ correctly
« on: February 28, 2014, 08:26:25 PM »
Hi, I will send this in to your support email but thought I would post here as well for others.

I am finding the latest versions of PS (1.01, 1.02) are not importing an OBJ exported from Mudbox correctly. It does not show any errors and displays the number of faces & vertices at the bottom of the window, but there is no geometry displayed at all.

If I use the 1.0 1760 build, it imports fine, though. I have also verified that other apps like Maya and Meshlab do not have issues importing the OBJ file correctly, either.

Cheers, Chad

Face and Body Scanning / Fast Avatar Capture
« on: February 13, 2014, 12:45:16 AM »

another Kinect-based setup, capturing and auto-rigging in 5 minutes to create simple avatars....nice going USC.

Bug Reports / Photoscan 1.01 Crashing on "Cancel" - OSX 10.9.1
« on: January 28, 2014, 10:03:49 PM »

I am finding that PS is crashing whenever I try to cancel an operation (even just loading a setup) in the 1.01 release build. It is working fine in the 1760 and 1795 builds. This is using MacOS 10.9.1 (Mavericks).

General / Aerial Reconstruction: How high, how dense?
« on: October 17, 2013, 06:17:06 PM »

This is for you aerial experts: What sort of limitations are there for reconstruction for very tall buildings and dense areas of construction, such as dense metropolitan areas with skyscrapers? I would imagine you would need lots of overlap and small increments to try to capture enough visual iterations down the height of a tall building or between tight spaces. Some folks around my office were curious, and I haven't really heard if there is a limit on what is possible.

I am sure flying limitations would also be a factor in many cases...this is just more of a theoretical query, and wondering what people have been able to achieve.

Thanks! -Chad

General / Strobe-based Image Projection
« on: September 11, 2013, 07:16:53 PM »
For those of you working with noise projections to help with capture, this seemed like an interesting concept.
Essentially it is a strobe lit slide projector.

Possibly it could allow a better capability to freeze motion than with the usual digital projector setups, or better synchronization and control with the rest of your camera/strobe setup.

Has anyone tried something like this?

General / More Mainstream by the Minute
« on: August 03, 2013, 08:11:39 AM »
For all of you 3D Photo Boothers:
They even handle pets!

I'm going to bring in a Puli and see how it goes:  ;-)

BTW, do you think their 3D printing process is using the same color reproduction technique as at Sculpteo?

It looks like they are getting somewhat reasonable reproduction of smaller details.

General / FreeD 3D Replay Tool
« on: August 01, 2013, 12:31:22 AM »

an interesting application of SfM

looks like they're only using six cameras in some cases:

and here's a video that shows a bit better how much detail they are capturing (or not) and some interaction with their system


A research project from Disney Zurich for those interested...

General / 2011 iMac vs OpenCL
« on: July 11, 2013, 09:45:44 PM »
Hi again,

I'm getting some mixed messages from PS when running it on a mid-2011 iMac (3.4 Core i7, Radeon HD 697M 2Gb). I'm running MacOS 10.8.4.

If I enable the Radeon in the OpenCL preferences, it does seem to contribute in a very small way to speeding things up, but in the console PS seems to be disabling it:

Using device: ATI Radeon HD 6970M, 12 compute units, 1024 MB global memory
max work group size 1024
max work item sizes [1024, 1024, 1024]
max mem alloc size 256 MB
selected 7 cameras from 7 in 0.00338 sec
Loading photos...
Reconstructing depth...
[CPU] estimating 1774x2277x288 disparity using 887x1139x8u tiles, offset -2
[GPU] estimating 2150x2317x576 disparity using 538x580x8u tiles, offset -379
GPU processing failed, switching to CPU mode

But further down in the log, it does appear to be using it in the build:

finished depth reconstruction in 514.051 seconds
Device 1 performance: 57.8953 million samples/sec (CPU)
Device 2 performance: 57.3216 million samples/sec (ATI Radeon HD 6970M)
Total performance: 115.217 million samples/sec
Generating mesh...
5467149 points extracted

Any ideas why this would be the case? Should I not be seeing a better boost from the GPU as well?
I've also run this on a newer iMac with the Nvidia GTX 680MX, and that one works wonderfully, almost cutting the geom build time in half.

Cheers, -Chad

General / Capturing a Coin and Merging Chunks
« on: July 10, 2013, 09:35:13 PM »
Hello again,

I'm trying to find an elegant way to capture a small coin, and have yet to find a reliable method that allows me to get the entire piece as a complete clean object.

If I shoot around it placed flat on each side and create two chunks in PS, I cannot get enough overlapping points to get the two chunks to merge reliably. This makes sense, as there is very little detail along the coin edges, and it's usually very uniform in appearance.

If I shoot it mounted vertically, I can usually get a decent build of the whole coin, except that I have to secure it somehow to keep it in that vertical position while I shoot it, and I would also like to see that part of the coin that is bearing the support. I have tried to shoot the coin vertically in two passes with the mounting point moved to a different location on the coin for the second pass; and I can get each chunk to build correctly on their own. But I cannot find a way to get PS to align the chunks correctly so that the second chunk with 'fill in the gap' created by the mounting point from the first chunk. Usually it still aligns the two chunks in the exact same shooting positions. I have tried masking out the mounting point in both chunks, but that doesn't seem to align at all.

I am only using the standard edition, so I do not have the option of defining correspondence points between the two chunks. But I would have thought that PS could still recognize enough overlapping information between the two chunks to figure out the alignment.

Are there any ideas out there on how to get this to work? I am stumped at this point.

Thanks! -Chad

General / Compact Canon CDHK RAW & Lens Distortion
« on: July 06, 2013, 08:03:44 AM »

I don't think this specific question has been answered before, apologies if it's been covered already.

If I am using CDHK with a Canon compact camera (A3300 IS) and shoot in DNG raw, it does not automatically correct for the lens it appears different from the JPGs coming out of the camera. With most cameras this is the expected behavior, aside from some cameras that bake in the distortion correction into their RAWs as well.

Will Photoscan still calibrate correctly for these DNG versions from this camera? I would assume that I would not want to apply any distortion correction in another program before sending it to Photoscan, right?

I know Photoscan should normally handle this as expected, I just wasn't sure in this case, since this camera normally can't shoot RAW.

I'm attaching jpg/dng versions at 28 and 51mm to show the difference between the two versions. Thanks for any info. I imagine it will probably be fairly obvious when I test it out anyway.


General / Getting Holes...Need More Photos?
« on: July 03, 2013, 09:13:09 PM »
First time exciting! I just want to start with a huge appreciation to all of you who answer questions so readily in this forum, it is an amazing resource for a great piece of software. So far I've made it through about half of the messages in the forum, lots of helpful info!

I've just starting getting familiar with SfM and Photoscan and have been experimenting with a variety of shooting styles and subjects, mainly with an emphasis for archeological presentation.

I photographed this sarcophagus at a local museum recently, but I am getting some holes and other artifacts in the texture (see right side in screenshot), no matter what method I use for generation (Average, Mosaic / Generic, Adaptive Orthophoto, etc). The geometry is generally coming out okay, but if I choose a higher quality mesh, I start to lose some feature detail, and I'm getting better results with low quality. I shot raw and have boosted the shadows quite a bit, so there isn't much that is dropping into darkness at this point. The photos are a bit noisy, so that could be creating some issues with the solve.

The camera alignment also seems to be okay for all of the photos, as you can see in the screenshot.

My guess is that I need take more photographs from a 3/4 angle along the main length of the sarcophagus in order to get more visual data for the 'nooks and crannies'...and probably try to work at a lower ISO to get the noise down a bit more.

Does that seem like an accurate assessment? Any other thoughts from the brain trust out there?

Cheers, -Chad

Pages: [1]