As many users point out before, those “red line”, “grid pattern”, “artifact” occur when the image is process by GPU.
If disable GPU and use CPU, those artifact will be disappear.
GPU:
https://i.ibb.co/9gzZgq9/GPU.jpg
CPU:
https://i.ibb.co/Z6Nrpkt/CPU.jpg
Side by side comparasion:
https://i.ibb.co/ysq5JFW/2020-07-04-121654.jpg
AIE vs GP-GPU vs GP-CPU:
https://i.ibb.co/8cvGdcW/2020-07-04-124102.jpg
2 Likes
That’s not my computer fault. Check lhkjacky post.
1 Like
Can confirm the findings.
Good that my new computer is already faster with CPU OpenVINO than using my GTX1060 graphics card 
The AMD 3900X is a real beast for such tasks.
Hope they will retraining the GPU AI models new to get rid of this ‘bugs’
1 Like
I can confirm the problem of GPU mode giving these weird horizontal lines, but not when using CPU. Which in my case is much much slower. I have seen this behavior on and off for last few versions. Please fix it.
Application & Version: Topaz Gigapixel AI Version 5.0.1
Operating System: Windows 10 (10.0)
Graphics Hardware: GeForce GTX 1060 6GB/PCIe/SSE2
OpenGL Driver: 3.3.0 NVIDIA 451.48
CPU RAM: 16266 MB
Video RAM: 6144 MB
Preview Limit: 7229 Pixels
2 Likes
I went one step further and found that it only happens with GPU when using the Max Quality models in Natural mode.
That particular image has pretty heavy banding in the jpeg artifacts, and I wondered if that has anything to do with the lines showing up. I ran it through DeNoise AI with default settings to clean up the compression artifacts and voila, no more GPU lines.
Application & Version: Topaz Gigapixel AI Version 5.0.1
Operating System: Windows 10 (10.0)
Graphics Hardware: GeForce RTX 2070 SUPER/PCIe/SSE2
OpenGL Driver: 3.3.0 NVIDIA 442.74
CPU RAM: 16312 MB
Video RAM: 8192 MB
Preview Limit: 8000 Pixels
2 Likes
For me that is also true., GPU enabled + Natural Mode. You can probably get rid of lines if you increase the remove noise slider in GigaPixel but that retains less details during up scaling so its not a good solutions. Its a bug that needs a fix.
Looking in the dark area under her chin I got the most complete removal of jpeg artifacts with the Max Quality, GPU, Man Made combination. The other combinations all had some residual red hash marks.
Man Made also does a nice job of sharpening her eyebrows, but I’m not crazy about having that much detail show up in the whites of her eyes for a portrait.
I believe your drivers are out of date, the latest is 451.48 available from the NVIDIA website.
1 Like
Those are targeted at Win 10 2004 which I have not migrated to yet so I was in no rush to update drivers. I went ahead and updated the drivers anyway, and as I expected the results were pixel for pixel identical to the results I had an hour ago with the older drivers.
This version still has serious color problems. As long as the image appears black and blue, the image export will appear reddish. It works normally in version 4.4.5. Even if the preview will appear red, the export works normally. This is a Very serious problem, hope to solve it.
https://i.ibb.co/QnNGsKQ/test.png
https://i.ibb.co/ydFv3dp/test2.png
1 Like
Hey Andy, I made the image highly compressed to test how good Gigapixel was at removing the artifacts. I figured it was a good test of AIE as well. Thanks for your test info.
Ron
Huh? OpenVINO is an Intel software that uses an Intel CPU feature:
" OpenVINO toolkit (Open Visual Inference and Neural network Optimization) is a free toolkit facilitating the optimization of a Deep Learning model from a framework and deployment using an inference engine onto Intel hardware."
@lhkjacky I tried it myself and you are certainly right, the CPU (even though it is slower) with Open VINO produces a better results. I didn’t find any difference between the Max Quality setting on or off when using the CPU. With MQ off, it will run approximately twice as fast. I also hope Topaz can improve the GPU results to be like the CPU. I assume they both use the same model.
Just saw tjwmmd2 reply on Open VINO. I have an AMD 2600X CPU and Open VINO works and speeds up the process.
2 Likes
But that makes no sense though.
The supported hardware for OpenVINO are given below
- 6th to 10th generation Intel® Core™ processors and Intel® Xeon® processors
- Intel Pentium® processor N4200/5, N3350/5, or N3450/5 with Intel® HD Graphics
And this post:
I dunno, seems like parts of OpenVINO that use SSE2 and doesn’t use the Intel optimizations may work on AMD.
But it doesn’t mean that it doesn’t work.
Look in the log files:
2020-07-04T11:35:06.293 [Debug ] <TEUpsamplingWorker::procMat line #: 441> Thread ID: 0x1f69796c2d0 ===========================================================
2020-07-04T11:35:06.293 [Debug ] <TEUpsamplingWorker::procMat line #: 442> Thread ID: 0x1f69796c2d0 Start image processing
2020-07-04T11:35:06.293 [Debug ] <TEUpsamplingWorker::procMat line #: 443> Thread ID: 0x1f69796c2d0 ===========================================================
2020-07-04T11:35:06.293 [Debug ] <TEUpsamplingWorker::procMat line #: 445> Thread ID: 0x1f69796c2d0 !!!!! CPU
2020-07-04T11:35:06.293 [Debug ] <TEUpsamplingWorker::procMat line #: 461> Thread ID: 0x1f69796c2d0 !!!!! OpenVINO
2020-07-04T11:35:06.293 [Debug ] <TEUpsamplingWorker::procMat line #: 462> Thread ID: 0x1f69796c2d0 - iGPU available (m_igpu_available): false
2020-07-04T11:35:06.293 [Debug ] <TEUpsamplingWorker::procMat line #: 463> Thread ID: 0x1f69796c2d0 - Enable iGPU (m_enableIGPU) : true
2020-07-04T11:35:06.293 [Debug ] <TEUpsamplingWorker::procMat line #: 471> Thread ID: 0x1f69796c2d0 Model file: "C:\\ProgramData\\Topaz Labs LLC\\Topaz Gigapixel AI\\tgrc\\giga-ov.tz"
2020-07-04T11:35:06.293 [Debug ] <TEUpsamplingWorker::procMat line #: 519> Thread ID: 0x1f69796c2d0 !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
2020-07-04T11:35:06.293 [Debug ] <TEUpsamplingWorker::procMat line #: 520> Thread ID: 0x1f69796c2d0 ----- OPENVINO FP32 "C:\\ProgramData\\Topaz Labs LLC\\Topaz Gigapixel AI\\tgrc\\giga-ov.tz"
It use the OpenVINO FP32 AI model from topaz.
According to Wikipedia, AMD Ryzen CPU support instructions sets “SSE1, SSE2, SSE3, SSSE3, SSE4a, SSE4.1, SSE4.2 and also AVX, AVX2,”.
I am using AMD Ryzen 7 1700 cpu, enable OpenVINO in GP speed up the CPU process significantly.
1 Like
I noticed that using CPU with this version I only get 40% of CPU usage when processing images.
I still have v 4.4.5 and it gets 60% of CPU usage. And of course it was much faster with v 4.4.5
CPU : AMD Threadripper 1950X 16C/32T
1 Like
Can confirm this finding.
On my 3900X 12c/24t the version 4.x got 80% CPU usage the 5.x only 50% max.
Maybe Topaz use a new version of OpenVINO which is more optimized on Intel and less AMD ready, but for my usage I still prefer OpenVINO over GPU model.
I have a Ryzen 2700x and with the OpenVINO I get worse results than CPU mode only. More noise, less sharper and more glitches.
CPU vs GPU vs OpenVINO:
https://i.ibb.co/CHKsG8z/2020-07-05-173827ss.jpg
Yes, the CPU version is slightly sharper than OpenVINO, but both of them are much better than process by GPU. Those “Red line” or “Grid Pattern” only show when using GPU.
In my old Ryzen 1st generation 1700 processor,
Process by CPU took 100 seconds at 70~80% cpu consumption.
With OpenVINO enabled, it only took 40 seconds at around 30~50% cpu consumption.
While using GPU, it took 20 seconds.
Processing Time:
CPU (100s) > OpenVINO (40s) > GPU (20s)
