same here, impressed of frigate and getting some great shots sometime 
Having run the past versions, I believe this is mostly to do with the fact that its recognizing more stuff in general. My false positives on animals have scaled up about the same as my correct positives on humans. Iāve been able to weed most of them out with min_size though Iāve had to go a bit larger on that than I want (ie can be some edge cases for people sizes too).
Is there any reason the PCI version of the coral wouldnāt work?
@scstraus @nic0dk @AdamWood I started an issue here if you want to upload any sample clips of animals that are false positives for humans.
2 Likes
Iām curious if anyone else has had the same issue as me when running any of the rcās of 0.5.0 or 0.5.1. I have a small Dell Optiplex with an i7 processor that has run Home Assistant, Frigate, and other containers now for over a year. I had been running Frigate 0.4.0-beta for a good while without any issues. When @blakeblackshear released 0.5.0, I upgraded and it seemingly ran fine for a day or two. But one morning I woke up and noticed that the entire machine was frozen solid, and the Coralās white LED was fast-blinking. I wasnāt sure what happened, so I rebooted and went on with my day. However, the same thing ended up happening a couple of days later as well.
The only change I had made to the system was upgrading Frigate from 0.4.0-beta to 0.5.0-rc7. I tried a third time and upgraded it to 0.5.1-rc1, but ended up with the same hanging issue. Iāve since reverted to 0.4.0-beta about a month and half ago (the one that had been running without issues for a long time) and the hanging issue has not returned.
From what I could tell based on some Home Assistant sensor history, there wasnāt any major CPU or memory usage at the time of the crashes. Frigate is set up with three cameras and all is back to normal again on my old 0.4.0-beta Docker image.
Anyone else experience anything similar?
I need to push up a final version for 0.5.1. I think all the issues are resolved now. You can use the dev image which should contain the latest fixes.
2 Likes
Do you think the dev version fixes the hanging issue I had?
It is very likely.
Great. I pulled the dev container and itās running now. Iāll report back if there are any other issues. Thanks!
Latest release is super solid. From what Iāve seen so far, it could even be a 1.0 RC.
So I realized I forgot to include a shm-size parameter when I moved from 0.4.0-beta to the latest dev container. Is there a default value that gets used without specifying anything on the command line or in docker-compose.yaml? All seems to be working fine without specifying it on my 3-camera setup.
EDIT: The Docker docs say that 64M is the defaultā¦ I could bump that up and manually specify the size if necessary, but Iāve been getting notifications of detections since this morning when I migrated to dev. Should I increase it or leave it?
That gives the container more room to store frames in shared memory. It may skip some frames with the default value. It just depends on frame rates, resolutions, and what is happening across all your cameras at the same time.
1 Like
Was hoping for a quick sanity check before I grab another coral device. Currently running an esxi host with a USB controller passed through to an ubuntu virtual machine running docker and a coral device on it. Im running 2 8mp cameras and 2 3mp cameras. The problem im running into is high detection rates (40-60 fps) on the cameras which put the coral device around 100fps. Is this normal to have such high detection rates? I have a lot of trees which create shadows which I suspect to be part of the issue. i have all my cameras set at 5fps max. The vm is also running in the 10-12ghz range while the detection rates are high. Thanks for any input!
Have you configured masks to avoid motion detection from triggering in areas you dont need to look for people?
I do have masks setup, it has not seemed to help much. Is there any possible miss configuration that could make the detection fps high?
If you really have that much motion in areas where you want to watch for people, you will see high detection fps. It is windy where I am today, and my coral is running at about 60-80 fps. That drops down substantially at night. I may improve the motion detection in the future, but for now frigate will end up skipping frames to compensate.
Im guessing I may need to bump a couple of my offending cameras down to their 640x480 steams for now. Its pretty windy here as well and just 1 camera has detection of 85fps which has the coral at 120fps. Thanks for the reply, im a noob and figured i set something up incorrectly.
I am getting detection of a flower basket hanging on my porch after dark. Any chance to eliminate this?
55% is a pretty low score. You can bump up your threshold a bit or add a black area in your mask for the bottom of the bounding box.
Iāve found 88% to be the magic threshold for accurate human detection. Iām sure others have decided differently, but 55% I think will be way too low for any normal use.