CPU hardware requirements for medium system

[SYPJ]

I currently have a similar thread going regarding CPU usage, and so far it's looking to me that to run six, 3MP cameras, you're going to need some serious hardware. My current setup for testing has four 1080p feeds, and that looks to be the max with this hardware, which is no slouch. CPU is an 8 core AMD FX-8350 at 4.0 GHz.

-Jason

I'm currently using different software that does do analysis (not as extensive as sighthound), but allows me to do the analysis on the low resolution stream of my cameras and record the high res stream, so it's CPU use is fairly low. That said, I pulled down the Sighthound trial and installed it on my server which is a Opteron 4280, similar to the FX-8350 but higher latency in the cache and lower IPC (Bulldozer vs Piledriver cores) and a base frequency of 2.8 GHz. I added 4 cameras and kept the resolution native, 3 of which are 3 Megapixel and 1 1080P. Based on your experience with the 8350, I expected this to be a complete no-go. Surprisingly, the server is running along typically between 75-85%, and will max out at 100% for a few seconds when I export a clip.

I'm curious why the different results, perhaps you have much busier scenes and more analysis going on with your cameras? I'm just running with the default rule to test. I do have my own FX-8350 desktop that I'm curious to try this with, but I already put the trial on my server, so I have two questions for someone at Sighthound:

1st, Can I run the trial on my FX-8350 also to test and see how it handles the load? If I decide to switch to Sighthound it's clear the Opteron 4280 won't cut it with all my cameras (9). My personal preference for CPUs is AMD rather than Intel, so the FX-8350 should give me an idea of what I can expect if I build a new server out, other than that the 6300 series Opterons have double the layer 3 cache.

2nd, Sighthound was very impressive the first day I tried it, admittedly it was cloudy. Day 2 was sunny, and while far less false positives than my current software, it still identified shadows on my lawn and patio as people once in a while. What kind of accuracy should I expect if I decide to purchase, or can I possible tune it a bit?

Thanks for your time and help!

BTW, count me in on voting for the ability to zoom when looking live or at clips. That's almost a deal breaker, but I will be willing to trade that feature if the "people" detection can cut down on the false positives by 90%. I did have a few cases of people in plain view it did not record which was also confusing; my other software did detect them and record. Does Sighthound need them to be moving to detect them?

[ryan]

CPU usage will be higher in busier scenes as object tracking and classification do require additional processing over an empty scene.

We'd be happy to get you a trial for your second machine, however it appears that the email used for your forum account is not the same as for your installations. Please let [email protected] know the correct address and we'll get you set up.

There are no knobs to turn in the software for recognition accuracy, we strive to have a zero configuration, works in all situations solution. The paid product does not differ from the trial, so you can expect identical functionality and accuracy.

Our computer vision lab has been hard at work on our next generation algorithms which we hope to integrate into the product soon. While these will not fully eliminate false positives, they should demonstrate improved results for many users and scenarios. Even more importantly they provide a better platform on which we can both improve and expand functionality vs. the implementation currently in the product which is pushed to its functional limits.

While computer vision is certainly a key differentiator to our product, we certainly aren't ignoring other areas of the software. Performance is very high on our list, particularly with increasing resolutions and decreasing costs. We've come a long way from the days of 4 640x480 cameras being the expected maximum. We already have a few significant performance improvements implemented and under testing, and plans and dreams for many more to come. We'd love to have your disk I/O be the first performance barrier you hit, not your CPU.

Note that the above is not a pitch to sell the software, but more of a heads up that we are indeed hard at work on the software and numerous improvements are planned for the near, medium, and long term. If you're not yet a current Sighthound Video customer please take full advantage of the pro trials to evaluate. We certainly hope it wows you in its current form, and if it doesn't we hope to win you over as a customer with a future release.

[Walt Dockery]

I'm currently running three 1.3MP 720p IP cameras off of a mac mini (late 2014, 2.6Ghz, 8gb ram) and it uses about 26% of the CPU, I will be adding one more 1.3MP and two 3.0MP cameras in the future, I will update and let you know of the CPU usage.

Any update as far as how the mac mini is handling the additional cams?