 |
|
|
|
|
|
 |
| ||||||||
 |
 
|
|
|
|
|
|
|
|
|
| ||||||||
|
|
| 
| User list | You are not logged in.
I`m on a fresh installed medium hi spec Win 7 with 6 Gb ram using Gigapan 2.06
When using smartblend to render a 4 gig 996 images pano it takes forever. Whats normal expected rendering time for smartblend on 2-4 Gpx panos and whats the bottleneck ? CPU usage is low and Ram usage is modest most of the time.
Offline
To give you an idea:
3 x 14 Panorama = 42 images at 88mb per tiff file (psudo-hdr in photomatix)
Laptop HP Pavilion dv8000t, 2.16 GHz Intel core duo, 2 gb ram, Windows XP sp3, autopano 2.0.5.
Render time =3 hours, Smartblend time = 7 hours. Total = 10 hours.
117 links between images, 6,515 control points.
Size = 29028 x 10388, FOV = 143.74 x 48.97, RMS = 2.47, Lens =Std, Color = LDR, Projection = spherical.
Final Output size =2.355 GB tiff image file.
I tried on my new macbook pro, osx 10.6.2 combo, similar pano and crashes at smartblend stage. I can not use my hp laptop anymore beacause it is malfunctioning.
Manuel.
Offline
Compare that to Multiband and you'll be amazed 
Smartblend is slow, because it does not depend ot CPU or RAM. It depends on HDD mostly. I advise you get a decent SSD drive for temp folders or at least build a RAID0 out of WD Black series drives. Or use Multiband.
Or be patient By the time your render is over it's quite possible that Alexandre will be ready with that MagicBlend logic that works better than Smartblend and Multiband combined AND uses CPU+RAM+GPU.
Offline
[bo] wrote:
By the time your render is over it's quite possible that Alexandre will be ready with that MagicBlend logic that works better than Smartblend and Multiband combined AND uses CPU+RAM+GPU
GOOD-NEWS!! 
best, Klaus
Offline
Forget smartblend for a gigapixel picture, multiband is a lot faster for the same result. There is no need to use smartblend because you have a lot of pictures. Moreover, smartblend uses a lot of space on your DD (many Go, and for your picture I thought it would be 1To).
Offline
Thank you Bo for the news and tips, i am happy todate with smartblend, as I am only stitching small panos for my hobby, but I will also try multiblend. I will just wait for the next update of APP to include probably the magicblend or I may upgrade to APG + APT + KRPano someday soon, when APT become fully standalone.
Manuel
Offline
manuel rosario wrote:
I may upgrade to APG + APT + KRPano someday soon, when APT become fully standalone.
Manuel
APT is and always has been a separate standalone prgram. It has just not been sold separately but only as a bundle with APG.
Offline
Thank you Andrew for the explanation, I hope someday I could advance to the stage where you are now. I wish I could do some virtual visits (tours). I am only at the APP (panorama stage) at the moment.
Manuel
Offline
With 996 images theres too much ghosting in my pano when using multiband, Its a cityscape with cars, people etc, and sea with boats. With the multiband version I`ll have to photoshopp to many ghostings to get a decent result.
I have a render only computer so it might render for days if it has to, but it seems to freeze half way through with very little activity. ( or crash )
Is it possible to render the pano as 4 seperate parts and then stitch it together ? ( Smartblend works OK for images up to 1 gig with my setup, but not with 4 gig )
I also have access to a renderfarm used for 3DSmax film rendering, if I could figure out how run smartblend on it.
Last edited by JohnM (2009-11-26 19:51:16)
Offline
JohnM: yes, you could crop the pano to 1/4 and render each part separately, then combine in postprocessing. I would leave a tiny overlap between the crops, just to be safe, in case there's a bug in the crop code and you end up with 1 column of missing pixels. 1 row of pixels on a gigapan is not a big deal, and I never tried rendering such huge panos, but if there is a crop 'innacuracy', why take the risk :]
Offline
An overlap would be necessary to let smartblend do a proper job with colours/blending I guess.
I plan to use gigatiler to chop the final result into photoshop friendly tiles for manual photoshopping, but is there a way to revert the process ?
( btw: I`m really looking forward to the gigapixel editing tutorial. )
Offline
Your question regarded cropping up the pano because you couldn't render one large one. No point in cropping it if you plan on using AP to stitch the crops again. No need for overlap if there are no missing pixel rows/columns and you plan on combining the cropped parts into a whole in an image editor.
Offline
It came through with a render time of 48 hours.
Btw: I didnt plan to stitch the crops in Autopano, but smartblend would need the neighboor image info to do a nice job, would it not ?
Offline
JohnM wrote:
Btw: I didnt plan to stitch the crops in Autopano, but smartblend would need the neighboor image info to do a nice job, would it not ?
Not only smartblend but Autopano in general would need overlap to find common control points.
Offline
JohnM wrote:
4 gig 996 images pano it takes forever.
<warning "comment=APP newbie evaluating 64-bit trial" />
I've used the trial for panos up to 250 or so total files. For that situation, my practice has been to use PsCs4's ImageProcessor to generate half-or-third-resolution jpg's as a "draft" to see how the image looks before doing a "final".
From my Canon 50D with 4752x3168 (15mp), this is 2376x1584 (75% fewer pixels). These seem to be handled significantly faster (but I haven't done actual timings).
I've had more than several big panos be too flawed to be usable (such as gaps from insufficient or no overlap) .... seems better to find that out from the "draft" than have it grind for hours/days and then find out it's a reject.
Offline
I. Alexandre posted a great explanation about the challenges of the older smartblend algorithm here. I would advise you to use multiband for larger panoramas. They are working on developing a newer algorithm that is suited for the larger panoramas.
http://www.autopano.net/forum/t7850-sma … ion-sum-up
II. The multiband misalignment may be due to two problems.
1. The alignment for the exposure brackets may not be good especially if you used the Photomatix MTB method for brackets that are beyond two stops from the overall exposure. Examine the pre processed exposure brackets for misalignment.
2. The multiband blending algorithm may have a problem with the high frequency band pass filters. Bandpass filters for multiple T transition zones are used to solve the luminance blending ( larger T ) and the alignment issues ( smaller T ). I have especially noticed this with alignment of trees or other fine features. ( not bracketed, single exposure )
( Adelson, A Multiresolution Spline w/ Application to Image Mosaics, ACM Transactions on Graphics, Vol 2. No 4, October 1983 )
III. I would be interested in feedback about how much memory is useful for larger panoramas. The newer chipsets allow for 16GB of DDR3 w/ 2 channels ( Intel 1566 and AMD AM3+ ) and 24GB of DDR3 w/ three channels ( Intel X58 for 1366 socket ). Currently one of the computers I use supports up to 8GB with two channels. ( A reminder that the memory controller has been integrated on to the CPU on Intel as AMD integrated it earlier on to reduce latency to system memory. So memory is dependent upon the particular CPU )
IV I really appreciate peoples feedback about experiences with SSD and Western Digital Velociraptor 300GB 10K drives.
V. Unfortunately the short term Intel roadmap does not integrate a higher performance GPU on the CPU die. It is a lower end GPU.
Last edited by Castillonis (2010-01-26 20:05:46)
Offline
Castillonis wrote:
III. I would be interested in feedback about how much memory is useful for larger panoramas. The newer chipsets allow for 16GB of DDR3 w/ 2 channels ( Intel 1566 and AMD AM3+ ) and 24GB of DDR3 w/ three channels ( Intel X58 for 1366 socket ). Currently one of the computers I use supports up to 8GB with two channels. ( A reminder that the memory controller has been integrated on to the CPU on Intel as AMD integrated it earlier on to reduce latency to system memory. So memory is dependent upon the particular CPU )
And the Xeon 55xx series can have 192G in a dual-processor configuration.
Offline
