Gigapixel AI v4.9.2

nope, all it says “4 posts in topic”

@AiDon Can you give Kiko the ability to post links or upload pics? Thanks!

@kikogod done Taylor, able to upload images now using the arrow.

1 Like

Is anyone else getting longer processing times with the last update? I am on the trial version right now and was about to make the purchase, but today’s update makes the enlargement process take way too long for me to justify the expense. The same images I tested with yesterday took less than a minute to increase to the selected size. Today - using the same size increase, the process is now taking almost 24 MINUTES to complete per image.

Hi Patricia,

Can you try going to File → Preferences → “Use recommended settings” and try processing again? Any performance improvement?

Thank you. I had not seen that option before until a few minutes before I saw your reply and was already giving that a shot. The results were much faster that go around.

Glad to hear! Some updates will reset that setting unfortunately - if it seems particularly slow then give that a go.

If you have any other issues feel free to post here or send me a PM. Thanks!

1 Like

I’ve experienced constant crashes with the new update. I have to force-quit on every process. I did the “recommended settings” and it calibrated. I have the newest iMac with 40gb of RAM and the fastest iMac processor at 3.6Ghz. The graphics card is the Radeon Pro 580X at 8gb.

The result after calibration is:

  1. No on Discrete GPU
  2. Low on graphics processor use
  3. Yes on Intel VINO

How is it possible that I can’t use this software and my GPU? The only way to get a file processed and not have it crash is to turn off the GPU. Processing a small file to 2x increase is slower than on my MacBook Pro from 2013. Any ideas?

Hi @AiDon,

Now after testing on a similar environment like yours and going through the log file, finally I found what causes this result.
We tested on this spec of the machine:

  • Intel Core i5-7300HQ CPU
  • Intel HD630
  • Discrete GPU: GeForce GTX 1050/PCIe/SSE2 (Video RAM: 4096 MB)
  • CPU RAM: 8013 MB

First, this is the benchmark result:

  • GPU processing speed = 11.38 sec/megapixel
  • CPU TF processing speed = 100.04 sec/megapixel
  • CPU VINO processing speed = 22.92 sec/megapixel
    ( Actually not CPU VINO, that is OpenVINO iGPU if exists iGPU )

From the benchmark result, calibration set up the options
: Enable discrete GPU (YES) & Enable Intel OpenVINO (YES)

We run an image (1920 x 1080) including 13 faces with 4x with Face refinement (ON)
And I checked all OpenVINO module runs on iGPU, not CPU.

  1. Enable discrete GPU (YES) & Enable Intel OpenVINO (YES)
  • Main model (GPU) : 29 sec (Process time)
  • Face model (OpenVINO) : 1 min 23 sec (Loading time) + 0 sec (Process time)
  • Total : 1 min 52 sec
  1. Enable discrete GPU (YES) & Enable Intel OpenVINO (NO)
  • Main model (GPU) : 29 sec (Process time)
  • Face model (TensorFlow CPU) : 2 sec (Loading time) + 13 sec (Process time)
  • Total : 44 sec
  1. Enable discrete GPU (NO) & Enable Intel OpenVINO (YES)
  • Main model (OpenVINO) : 1 min (Loading time) + 2 min 31 sec (Process time)
  • Face model (OpenVINO) : 0 sec (Loading time) + 0 sec (Process time)
  • Total : 3 min 31 sec
    !! Loading time in OpenVINO Face model is 0 sec because we have already loaded it by #2 test.
  1. Enable discrete GPU (NO) & Enable Intel OpenVINO (YES) - (2nd round with OpenVINO in order to excluding OpenVINO loading time.)
  • Main model (OpenVINO) : 0 sec (Loading time) + 2 min 31 sec (Process time)
  • Face model (OpenVINO) : 0 sec (Loading time) + 1 sec (Process time)
  • Total : 2 min 32 sec

We have already known the fact:

  1. OpenVINO loading takes time.
  2. Especially OpenVINO iGPU loading takes a much longer time than OpenVINO CPU.
  3. OpenVINO is faster than Tensorflow CPU when considering only processing time.

From this test result,

  1. Intel OpenVINO iGPU: Loading time takes much longer than what I expected (1~1.5 mins)
  2. The second time for OpenVINO saves loading time.
  3. Even if the calibration result shows GPU is 2x faster than OpenVINO iGPU, GPU is 5x faster than OpenVINO iGPU in the real image processing time.
  4. GL/GPU module loading time is < 1 sec.
  5. Tensorflow module loading time is also small (2 sec.) but the processing takes longer. (15 sec. for 13 faces)

So calibration suggests Enable discrete GPU (YES) & Enable Intel OpenVINO (YES). I think that the first running process, the best option was 2) Enable discrete GPU (YES) & Enable Intel OpenVINO (NO): 44 sec, but after running processing a couple of time with OpenVINO, and also if there are many faces detected in the image, the options (Enable discrete GPU (YES) & Enable Intel OpenVINO (YES)) will be the best.
I guess OpenVINO CPU loading time doesn’t take as long as OpenVINO iGPU.

Obviously, there are somewhere to be improved. Especially these options, Enable discrete GPU (NO) & Enable Intel OpenVINO, make confusing everyone, and calibration is partially inaccurate.
And still, we need to investigate what causes OpenVINO multi-block processing longer than a single block processing.

Best,
Myunghoon

1 Like

Hi,

OK that reply, thanks for the details description, gave me another idea and that was to use the DCH drivers instead of the OEM drivers. As it was there are no customisations by HP to the drivers but the OEM locks meant that you couldn’t install DCH drivers previously.

In this particular case, the one with Face ON and the 6 faces in the 2400 x 1200 scaled up by 2x, the processing time dropped from 2:06 mins to 1:06 mins, using the both GPU & OpenVINO options set to YES, similar to in the previous version of GigaPixel AI.

But the most significant change is that the OpenVINO only processing, that took more than 4 minutes previously, only took 1:44 minutes this time. Just a little more than GPU + OpenVINO.

So, in summary, the DCH drivers seem to make a big difference for processing Faces and, in the case of the image I tested, the processing time was dropped to equal that of the GPU + Face.

And, as you said above regarding the speed of the Tensorflow module, I found that the processing time went from 1:13 mins to 1:27 mins … ??? I guess it could be because of loading times etc.

But, for the second example of the image with no faces, the following times were experienced:

OpenVINO Only - 1:41 mins (instead of 3:53 mins)
GPU Only - ** 1:11 mins** ( instead of 1:06 mins)
GPU + OpenVINO - ** 1:10 mins** (instead of 1:05 mins)

Again the DCH drivers seem to make a big difference for OpenVINO only processing.

But there are inherent issues in using DCH drivers for iGPU because of possible OEM customisations and I am wondering if that negated the iGPU inclusion in the OpenVINO only processing and the DCH drivers opened that up?.

Don

Hi, me again, trying to post the screenshot :slight_smile:

I’m seeing the same thing with the window frame trimming the edge of the menu. Resizing the window does not make a difference.

I’m using a custom scale factor (116%) to calibrate my font display size, so perhaps that has something to do with it.

Thanks for the screenshot Kiko! Are you using a custom scale factor like Andy is?

Hey itunes, sorry for the late reply. We actually have some great videos on our YouTube channel with a lot of the in-depth tutorials you’re looking for. Unfortunately we don’t have one for night shot mode in DeNoise yet, I’ll mention that to the team.

Unfortunately JPEG2RAW is not in active development at this time. We’ve been lightly exploring some options with it, but haven’t decided on anything at this point.

2 Likes

Also, I noticed that you have auto-detect settings checked but the slider bars are blue - do you remember the steps you took for that to happen? (it shouldn’t happen :slight_smile: )

I ran a quick check, and set the scaling back to 100%. Sure enough it resolved the menu edge clipping issue. My other reasons for needing the accuracy of having it set at 116% outweighs this minor annoyance, so I’ll live with it for now.

Ah thank you. Appreciate it. About JPEG2RAW. I thought it had great potential, but maybe instead of having one separate program to worry about, it would be great if you could add the technology you already have in existing programs. Maybe to Topaz Studio or to Sharpen AI. That seems like a logical place to add as another tab or something inside the interface to leverage already great technology. It would be shame in my opinion to let it just fall behind, since its a great restoration tool and can help save some shots that other programs cannot. Hope you do something with it., it was a great concept.

Cheers!

Is it about OpenGL implementation or should it be “OpenCL” term used in this post? That would clarify very much, cause Mac users would be happy :slight_smile:

I don’t know why in that particular time the sliders were blue, now they are yellow all the time. I noticed the guy above me has the same problem with the cut off interface and I want to elaborate that I am too with Windows scaling set to 125% because I have a 27 inch 1440p ASUS monitor and everything in 1440p is just too small if I don’t set up some scaling, so I don’t want to revert back to 100%. Also, I’ve had this scaling for a long time but didn’t have the cut off interface problem in the last few updates.

It is only OpenGL