I suspect that the DeepStack object detection isn’t a huge money maker, and I’m sure is a very time consuming project. However, it has been quite a while since the issues with the January release not detecting objects has been reported. If there is no intention to address these issues it would be nice if that was clearly stated. If it is just going to take a while that isn’t a huge problem because I can live with the September build that works if not optimally.
It would seem that it would be very easy to replicate the issue. Just do a clean Windows native install and feed it some pictures using HTTP. If there is difficulty replicating the problem I would be happy to come up with detailed scenarios, or try debugging the problem further. At the moment I’m out of ideas on how to track it down on my end though. I’d be happy to work on the issue with you if you have some suggestions on what to look for.
Personally, I only have one development machine. It is running the native Windows release with no VM, no Docker involved. I can try installing a VM I guess, but that is a somewhat major project. From what other people have said this wouldn’t be successful anyway.
For what it’s worth, here is my development environment: Windows 11, AMD 5950x, GTX 1660ti, 32gB RAM. I am currently running the September GPU build successfully. I am using the standard models for object detection so there is nothing unusual there. BTW, I do have face detection turned on, but face detection is only called on specific pictures, not my test pictures.
If you have a way or build capable of outputting a detailed debug spew I’d be happy to to send you everything you need. If really necessary I could even arrange for you to do some remote debugging if you have tool capable of that. In short, I’d be happy to help you get to the bottom of this.
I can attest that my custom deer model no longer works. Built-in models do, not sure if as well.
When I send my python send test image over, it gives the same success percentages, but I’ve had no confirmed hits via Blue Iris (cars, people, etc are ok) since installing the update. That’s been @3 weeks which hasn’t happened before.
Blue Iris snaps pics, but says “nothing found” and marks them as Cancelled. I have a couple of videos with deer strolling across the screen.
Apparently email notifications have now been turned on for this site. Yeah! Now I won’t have to come back here quite as often.
If you’ve had any app successfully identify any objects you are better off than me. I did a clean install with the January release, no modifications of any kind.
It appears that the author probably has other (money making) projects going on. His interest in DeepStack seems to come and go. Unfortunately his interest appears to have gone shortly after the (non-functioning, at least for me) January release.
I’m not sure why the January release is working for anybody using Windows native versions, but apparently it is. That is the big mystery. If it worked for nobody it would be easier to track down.
I have made as many offers to help run down this issue as I reasonably can. At this point I would suggest that as a minimum that the January Windows native release be withdrawn in favor of the functioning September release. That way anyone interested in using DeepStack Windows will get a fully functioning (for everyone) release. This is too bad, because I was looking forward to using the Threads capability.
Hi everyone, thanks for all the feedback on the January release. A major change that shipped in this release is the update to the PyTorch version to support newer GPUs, courtesy of @john’s great effort to make this happen. While this works for so many modern GPU hardware, it seems to have unintended impact of some more newer GPUs and some CPUs too, and some effect on Custom Model APIs as well, as reported in this thread.
Our team is a small team and we don’t have access to full spectrum of latest and recent GPUs to test all changes. But we are committed to fix this by testing on more devices as we can, perhaps work with members of the community as well who are willing to help in the testing to ensure we are able to publish soon a version that works for everyone, latest and recent GPUs and CPUs alike.
I agree, I installed 2022.01.01 on a fresh VM with a Nvidia T400 following the documentation on DeepStack’s page to the letter; I get various errors including Timeouts and on recognition with 2022.01.01.
However, rolling back to 2021.09.01 (unstalled 2022.01.01 and installed 2021.09.1 and kept all the CUDA and CuDNN libraries in place), DS started to work fine again.
I have no error logs in %appdata%\local\DeepStack for 2022, it just doesn’t seem to work?
It would also be good if older versions and documentation remain online, perhaps I missed it?
As previously mentioned, I am willing to test any builds you want tested. In addition, I am willing to setup a remote debug session if you are able to help me do that. That should be almost as good as having known incompatible hardware in your possession. Since mine won’t work in either CPU or GPU modes this would be an excellent test case. Remote debugging is often a pain, but it should be doable even across the world with the right tools.
It has been more than 2 months since the last post here, and more than 6 since the original release of the current (non-working for GPU) version… From a complete outsider POV, it doesn’t seem like supporting the latest CUDA version should be that much of a chore. Is there any movement on this? It does seem like the most damaging issue currently facing DeepStack.
It think most people running DeepStack would be installing a new system with new CUDA drivers. After seeing that DeepStack just doesn’t work for the current CUDA, most people would just move on.