Messages

1

General / Re: v1.3 Performance Regression

« on: March 07, 2017, 06:38:39 AM »

Thanks Alexey,

So I have to wait for a new 1.3 release or the optimization is ready in the Version 1.3.0 build 3772?

2

General / Re: v1.3 Performance Regression

« on: March 06, 2017, 05:44:49 AM »

Alexey,

Thank you, but the main problem is the merging is very slow, compared with a project with 2,000 images which was just a matter of 10 minutes, this took us around a week and eventually ate up all memory and could't finish.

3

General / Re: v1.3 Performance Regression

« on: March 03, 2017, 01:10:35 PM »

Any update with the merging problem ? My team is still pending and making no progress.

4

General / Re: Performance diff. GTX 1060 vs GTX 1070 in photoscan...?

« on: March 02, 2017, 06:42:26 AM »

FYI, we just got a 1070, some benchmarks for dense cloud generation, high quality, building sample

1.2.6

2017-03-01 04:44:06 finished depth reconstruction in 691.51 seconds
2017-03-01 04:44:06 Device 1 performance: 1174.18 million samples/sec (GeForce GTX 1070)
2017-03-01 04:44:06 Total performance: 1174.18 million samples/sec

1.3.0

2017-03-01 05:00:56 finished depth reconstruction in 208.065 seconds
2017-03-01 05:00:56 Device 1: 100% work done with performance: 487.714 million samples/sec (GeForce GTX 1070), device used for 343.541 seconds
2017-03-01 05:00:56 Total performance: 487.714 million samples/sec

5

General / Large sample dataset

« on: March 01, 2017, 01:40:02 PM »

As we will have large production projects in the future and we just built a powerful server it, we want to find some large dataset to do stress test with our stacks. But it seems it is quite hard to find on Internet, all I found so far are:

https://www.sensefly.com/drones/example-datasets.html
https://dronemapper.com/sample_data

I want to test some dataset as large as 10,000+ images, with geotagging and ideally with GCPs.

Is it possible to have larger dataset for test and training, or we can have other ways to build large datasets, e.g. combine all small datasets around the world into one/ dulplicating files/ buy datasets from providers?

6

General / Re: v1.3 Performance Regression

« on: March 01, 2017, 01:32:58 PM »

Alexey,

The project was saved in psx format with depth map.

7

General / Re: v1.3 Performance Regression

« on: February 28, 2017, 01:46:58 PM »

Please help, we can't proceed anymore. 

8

General / Re: v1.3 Performance Regression

« on: February 27, 2017, 04:21:04 AM »

Please help!!

Now it will run for unbelievable time

http://imgur.com/a/4L9PJ

And all 128GB ram is eaten up !!!

http://imgur.com/a/5neeQ

9

General / Re: v1.3 Performance Regression

« on: February 24, 2017, 09:07:00 PM »

24MP, multi-spectral image pairs

Other facts: http://imgur.com/a/jItfz ( those 3 expanded chunks)

While at the meantime, the estimated merging time keep growing http://imgur.com/a/n1BMi

 

Does it matter that we enabled "save depth map" before we start chunk merging?

We had quite a quick result with v1.2.6 and around 1000 images dataset before.

Thanks

10

General / Best practices for handling 30,000 + images

« on: February 24, 2017, 11:44:29 AM »

Hi all,
We are going start a aerial survey project which will have around 30,000 images collected. We will use Agisoft to process them and it is first time for us to handle such large amount of data.

So I want to ask for recommendation of both hardware and software sides:
- Recommended hardware configuration, ram, disk, GPU, CPU
- Cluster is necessary? If yes, how many nodes? The totally process time is reduced linearly to the number of nodes?
- Multiple chunks? How to balance number of chunks and the number of images per chunk.
- The best way to merge chunks? By markers? If yes, then we have to place GCP markers around the borders of flights?


Thank you in advance.

11

General / Re: v1.3 Performance Regression

« on: February 24, 2017, 11:30:46 AM »

Thank you Alexey,

As to my other questions (incredibly long chunk merging time and rolling back to 1.2), could you give me some suggestions?

12

General / Re: v1.3 Performance Regression

« on: February 24, 2017, 07:06:59 AM »

And "merge chunks"(6k images in total, 3 chunks, merge dense cloud and mesh) in 1.3 is incredibly slow

http://imgur.com/a/GdTal

13

General / v1.3 Performance Regression

« on: February 24, 2017, 07:01:47 AM »

Win 10 x64
Agisoft v1.3
Dual Xeon
R390x


We updated Agisoft from 1.2.6 to 1.3.0 and we observed performance regression on all our workstaions, e.g. for "build dense cloud",

1.2.6:
GPU: 269mil samples/sec CPU: 106 mil samples/sec
1.3.0:
GPU: 45mil samples/sec CPU: 15 mil samples/sec
 

Now we want to rollback to 1.2.6 but our projects are already saved in 1.3.0. Can we safely re-install 1.2.6 and open our projects?

14

General / "Not enough points to align" during aglin chunk

« on: February 23, 2017, 05:03:47 AM »

Agisoft v1.3
3 Chunks, in totally ~ 6600 images, all are aligned, dense cloud and mesh generated.

Align chunk:
Method: point based, Accuracy: low, Point limit: 80000, Preselect image pairs: enabled

Log:

Code: [Select]

17-02-20 16:57:28 photo 6653/1: 9989 points
2017-02-20 16:57:29 photo 6654/0: 9053 points
2017-02-20 16:57:30 photo 6654/1: 10195 points
2017-02-20 16:57:31 photo 6655/0: 9053 points
2017-02-20 16:57:31 photo 6655/1: 10408 points
2017-02-20 16:57:32 photo 6656/0: 9404 points
2017-02-20 16:57:33 photo 6656/1: 10638 points
2017-02-20 16:57:34 photo 6657/0: 9372 points
2017-02-20 16:57:35 photo 6657/1: 10805 points
2017-02-20 16:57:36 photo 6658/0: 9033 points
2017-02-20 16:57:37 photo 6658/1: 10500 points
2017-02-20 16:57:37 photo 6659/0: 8933 points
2017-02-20 16:57:38 photo 6659/1: 10479 points
2017-02-20 16:57:39 photo 6660/0: 8966 points
2017-02-20 16:57:40 photo 6660/1: 10558 points
2017-02-20 16:57:41 photo 6661/0: 8725 points
2017-02-20 16:57:42 photo 6661/1: 10097 points
2017-02-20 16:57:43 photo 6662/0: 8294 points
2017-02-20 16:57:43 photo 6662/1: 9787 points
2017-02-20 16:57:44 photo 6663/0: 7519 points
2017-02-20 16:57:45 photo 6663/1: 9171 points
2017-02-20 16:57:46 photo 6664/0: 7525 points
2017-02-20 16:57:47 photo 6664/1: 9392 points
2017-02-20 16:57:48 photo 6665/0: 7686 points
2017-02-20 16:57:48 photo 6665/1: 9606 points
2017-02-20 16:57:49 photo 6666/0: 7802 points
2017-02-20 16:57:50 photo 6666/1: 10151 points
2017-02-20 16:57:51 photo 6667/0: 7563 points
2017-02-20 16:57:52 photo 6667/1: 10022 points
2017-02-20 16:57:53 photo 6668/0: 7457 points
2017-02-20 16:57:54 photo 6668/1: 9646 points
2017-02-20 16:57:54 photo 6669/0: 7763 points
2017-02-20 16:57:55 photo 6669/1: 9540 points
2017-02-20 16:57:56 photo 6670/0: 8111 points
2017-02-20 16:57:57 photo 6670/1: 9614 points
2017-02-20 16:57:58 photo 6671/0: 8128 points
2017-02-20 16:57:59 photo 6671/1: 9976 points
2017-02-20 16:57:59 photo 6672/0: 7917 points
2017-02-20 16:58:00 photo 6672/1: 9594 points
2017-02-20 16:58:01 photo 6673/0: 7850 points
2017-02-20 16:58:02 photo 6673/1: 9312 points
2017-02-20 16:58:03 photo 6674/0: 7505 points
2017-02-20 16:58:04 photo 6674/1: 8924 points
2017-02-20 16:58:04 photo 6675/0: 7201 points
2017-02-20 16:58:05 photo 6675/1: 8906 points
2017-02-20 16:58:06 photo 6676/0: 7771 points
2017-02-20 16:58:07 photo 6676/1: 9122 points
2017-02-20 16:58:08 photo 6677/0: 7706 points
2017-02-20 16:58:09 photo 6677/1: 8837 points
2017-02-20 16:58:10 photo 6678/0: 7727 points
2017-02-20 16:58:10 photo 6678/1: 8691 points
2017-02-20 16:58:11 photo 6679/0: 7906 points
2017-02-20 16:58:12 photo 6679/1: 8937 points
2017-02-20 16:58:13 photo 6680/0: 8024 points
2017-02-20 16:58:14 photo 6680/1: 9340 points
2017-02-20 16:58:15 photo 6681/0: 7599 points
2017-02-20 16:58:15 photo 6681/1: 8837 points
2017-02-20 16:58:16 photo 6682/0: 6995 points
2017-02-20 16:58:17 photo 6682/1: 8507 points
2017-02-20 16:58:18 photo 6683/0: 7534 points
2017-02-20 16:58:19 photo 6683/1: 8809 points
2017-02-20 16:58:20 photo 6684/0: 8367 points
2017-02-20 16:58:20 photo 6684/1: 9416 points
2017-02-20 16:58:21 photo 6685/0: 8555 points
2017-02-20 16:58:22 photo 6685/1: 10121 points
2017-02-20 16:58:23 photo 6686/0: 9007 points
2017-02-20 16:58:24 photo 6686/1: 11118 points
2017-02-20 16:58:25 photo 6687/0: 9093 points
2017-02-20 16:58:25 photo 6687/1: 11405 points
2017-02-20 16:58:26 photo 6688/0: 9801 points
2017-02-20 16:58:27 photo 6688/1: 12599 points
2017-02-20 16:58:28 photo 6689/0: 9705 points
2017-02-20 16:58:29 photo 6689/1: 12887 points
2017-02-20 16:58:29 points detected in 10309.1 sec
2017-02-20 16:58:29 Selecting pairs...
2017-02-23 02:07:58 229420568 matches found in 205767 sec
2017-02-23 02:09:38 matches combined in 98.658 sec
2017-02-23 02:10:17 20102 of 20102 pairs selected in 0.011 sec
2017-02-23 02:10:17 Matching points...
2017-02-23 02:38:42 35817483 matches found in 1705.51 sec
2017-02-23 02:38:48 matches combined in 5.608 sec
2017-02-23 02:38:48 finished matching in 217929 sec
2017-02-23 02:38:48 setting point indices... 6529182 done in 9.29 sec
2017-02-23 02:39:04 generated 6529182 tie points, 2.68849 average projections
2017-02-23 02:39:07 removed 61750 multiple indices
2017-02-23 02:39:07 removed 3390 tracks
2017-02-23 02:39:35 adding 2443099 points, 513 far (10 threshold), 5 inaccurate, 708 invisible, 113 weak
2017-02-23 02:39:37 adding 1167769 points, 469 far (10 threshold), 6 inaccurate, 877 invisible, 50 weak
2017-02-23 02:39:39 adding 871752 points, 434 far (10 threshold), 2 inaccurate, 762 invisible, 47 weak
2017-02-23 02:39:39 Aligning groups by 299038 points
2017-02-23 02:39:39 iteration 0: 112330 points, 0.00360861 error
2017-02-23 02:39:39 iteration 1: 110267 points, 0.0034801 error
2017-02-23 02:39:39 iteration 2: 109738 points, 0.00345292 error
2017-02-23 02:39:39 iteration 3: 109584 points, 0.00344514 error
2017-02-23 02:39:39 iteration 4: 109535 points, 0.00344259 error
2017-02-23 02:39:43 Not enough points to align
2017-02-23 02:39:43 Not enough points to align
2017-02-23 02:39:45 Finished processing in 217987 sec (exit code 1)


It takes me 3 days to finish align chunks and after that I saw logs above. Does the line "Not enough points to align" matter? Do I need to run it with higher accuracy?

15

Bug Reports / Re: [Build dense cloud] Error: bad allocation

« on: February 21, 2017, 07:07:04 PM »

Thank you Alexey,

So network mode (even with only local node) is a recommended way to reduce memory consumption.

I have one more question about network mode, if I add one networked machine for acceleration, for example, one with 3 powerful graphic cards, then do I still need to install large amount of ram on it? Or less ram (e.g. 64GB) is enough ?