[Config Support]: Horrible Object Detection Accuracy #15419
Replies: 2 comments 15 replies
-
Since when I tried to submit this the 1st Time, Github threw an Error that the Body exceeded 65536 Characters (and DELETED everything I filled in the Form - thankfully I always save everything as a TXT file offline before sending 🤣) I need to put this in a separate Message/Comment Any other information that may be helpfulI had several Issues setting up Frigate in a (unprivileged) LXC Container under Proxmox, with the LXC Container refusing to start completely when I started playing with UID Mappings (which didn't really work anyways). Thus I installed Podman from source using my own Script and I am running Podman directly on the Proxmox VE Host as an UNPRIVILEGED User (user = There are mainly 3 Things that are not working well, besides the absolutely bad Detection Accuracy:
There's probably lots of stuff I copied from the Example Configuration from the Documentation (what is NOT recommended, but given that I couldn't find much else ...). I hope there is something very obvious and wrong with my Configuration that's relatively easy to fix, because as it is now, the System is even less reliable than the Hikvision object Detection sending an Email 😢. Thank you for your Insights 🙏. |
Beta Was this translation helpful? Give feedback.
-
the best way to debug detection accuracy is to use the debug live view. This will allow you to watch with bounding and motion boxes to see what Frigate is seeing. You can also screen record this so you can share it with us and we may able able to provide insights to what is going on.
you seem to be misunderstanding. Transcoding and object detection are two entirely different things. rocm is for object detection. If you just want to do transcoding then you can do that with your AMD GPU in any of the frigate docker images. |
Beta Was this translation helpful? Give feedback.
-
Describe the problem you are having
I am faced with a VERY inaccurate Detections off my several Hikvisions 4K and one 4MP Camera.
I am currently using the Hailo-8L Detector on 2 different Systems, with Image Tag
833cdcb-h8l
(ghcr.io/blakeblackshear/frigate:833cdcb-h8l
).Some Examples include:
CPU Usage is very low (5%-10%) in both Cases which is very good.
The Systems are configured as following:
A. System A
pve-manager/8.3.1/fb48e850ef9dde27 (running kernel: 6.8.12-3-pve)
) - Kernel6.8.12-4-pve
caused issues, that's why I'm using a slightly older oneB. System B
pve-manager/8.3.1/fb48e850ef9dde27 (running kernel: 6.8.12-3-pve)
) - Kernel6.8.12-4-pve
caused issues, that's why I'm using a slightly older onePodman Configuration is essentially Frigate behind a Caddy Proxy using IPv6 mainly with
snid
IPv4 Fallback for IPv4-only Clients (and to access 1 Camera that I didn't set up with IPv6 yet).There's probably something VERY wrong with the Configuration.
Hikvision Camera Substream Settings are as follorws: Resolution 1280x720p, Variable Bit Rate, Medium Picture Quality, 10 Video Frame Rate, 4096 kbps max Bit Rate, Video Encoding H.264, Profile Medium, I-Frame Interval 10, SVC Off, Smoothing 50
Version
0.15.0-833cdcb
Frigate config file
Relevant Frigate log output
Relevant go2rtc log output
Frigate stats
No response
Operating system
Proxmox
Install method
Docker Compose
docker-compose file or Docker CLI command
Object Detector
Other
Screenshots of the Frigate UI's System metrics pages
Any other information that may be helpful
No response
Beta Was this translation helpful? Give feedback.
All reactions