Inference speed frigate

All motion regions are scaled down to that size to run the inference. 5ms and 10. Jul 24, 2023 · The inference speed is between 9. Which baffles me because the cpu still does better than the coral. Jul 23, 2021 · Frigate doesn't run object detection all the time, but when it does, your CPU load should increase much less with a Coral. Can handle many cameras at 5fps depending on typical amounts of motion. I have two types of cameras. In different eras, the roles and capabilities of ships classified as frigates have varied. With my coral USB stick, however, only 20ms. My inference speed is around 37 in debug which i have no idea is good or not! Dave Edited July 29, 2021 by Flubster Spelling Aug 11, 2021 · I'm personally using eclipse-mosquitto. Maybe it's a problem with this server and how it handles the usb to pcie. The second one is the new unit. Within that VM I have docker-compose utilizing the coral via Frigate. Feb 5, 2024 · Reporting and inference control. video Mar 25, 2023 · I running on a NUC6 (I5-6260u) with 8Gb Ram. tflite a go, seems like the best accuracy vs inference speed, although as you say it might not relate to Frigate at all! I'll report back, I couldn't find many discussions on this topic, unless I am using the wrong search terms perhaps It is easy to add a mean line in Grafana. Dec 30, 2021 · Somehow I don't understand the inference speed representations under Debug. Jul 9, 2021 · Frigate is running as Docker container inside an Ubuntu virtual machine and Coral USB is mounted to the VM. You should have your fps set to 5 in your frigate config, and you should be using hwaccel args for ffmpeg if possible. And the object detection entities are also unavailable. It helps in reducing the time, computational resources, and cost required for making predictions, and can also improve accuracy in some cases. This tool is here to help calculate a frigate base stats. CPU and MEM process stats. My inference speed hasn't changed by much. 0-bfbacee. In my experience so far, anything using Google's FCM for push messages can be delayed. I've set min_score to 0. I am using optimal configurations defined for Reolink cameras on frigate wiki. Min-maxing frigates is not required to successfully Nov 17, 2023 · Configuration. 05 now from min_score 0. Award. It can be named frigate. Can leverage Intel QuickSync for stream decoding. 0 (ubuntu server 20. The USB stick is passed on to Hassio. 0-11babb95 This is a docker container that runs a Prometheus exporter for Frigate stats. Tested with 0. It worked, but inference speed was ~10x slower than with a PCIe coral (~100ms). Maybe it's an issue with how vsphere passes the pcie through to the vm. Inference speed is now a few ms slower but nothing in it really, it's still ~20ms. We have used techniques like inference computation May 8, 2023 · The difference is insane from pretty warm to just cold all the time. 13 (currently in beta) has some new motion detection settings like ignoring and recalibrating motion when large changes like lights, changing between color & IR, ptz moves, etc occur. Frigate provides the following builtin detector types: cpu, edgetpu, openvino, tensorrt, and rknn. So I've been self-hosting my CCTV for about 3 years now and it's always been not great. This does a rolling average over 30mins. xml model. 68. Frigate config file Apr 19, 2024 · April 19, 2024. do what was described at AMD GPU detector with yolov8 models #9446 (reply in thread) and move the model generation to S6 like tensorrt. Frigate continues to work well and I've now eliminated practically all false positives! Sep 12, 2023 · I am migrating over from running Blue Iris on a virtual Windows machine, and initially had great results with Frigate, but have been running into difficulties. Media Serving. 2) And the TPU is recognized: frigate. Barebones, not VM. Any reason for that or against using specific cores? detectors: cpu1: type: cpu cpu2: type: cpu cpu3: type: cpu Jan 3, 2023 · Please open the terminal app and type in following command: docker -v. Jul 12, 2023 · Describe the problem you are having coral P-ID Inference Speed CPU % Memory % 299 5062. What i'm dealing with right now is a delay of 10+ seconds between me doing something on camera and it showing up on the Frigate feed. 1 and 6. You can calculate the maximum performance of your Coral based on the inference speed reported by Frigate. When using multiple USB Accelerators, your inference speed will eventually be bottlenecked by the host USB bus’s speed, especially when running large models. 2 Frigate docker images. 04 and the driver is pre-installed and works fine on Frigate but you may wish to install the latest drivers. 0 and 3. Searching other posts show yolonas inference of 50-60ms at 320x320 but not sure if this is at idle or average. I did not troubleshoot or test this in depth, it looks like only USB2 speed was working for me rather than USB3, this is probably solvable, but I didnt try. At 320, my inference go up to around 220ms when motion is detected. Type in following command in terminal app to create frigate docker file with vi text-editor: 1 mkdir ~/Documents/frigate. I'm brand new and haven't changed anything, just the basic setting in the yaml. Running on VirtualBox on my 10 year old old IBM Ideapad y560p. 3 of a second it should detect them, if all other conditions are met. Oct 18, 2022 · I am running Frigate 0. BMAX B2 Plus: 10-12ms: Good balance of performance and cost. Inferences are run on a single size image. 14 beta although I suspect this issue still exists in ver 0. Need to add a new query dataset to the inference graph: avg_over_time(frigate_detector_inference_speed_seconds[30m]). So I recently added a second Google Coral usb. MDLefevere: remove the readme in the docker section and move that to the Frigate object detectors docs. Does this sluggishness seem about normal for Raspberry pi4 hardware, or is there likely a setup HTTP API. However the main difference between 2 fully leveled frigates are their traits. You seem to have problem using LXC, but it should the way to go if you are going to use Frigate with Proxmox. I have tried changing the detector ratios to 1280x720 and no difference except the detection seems to be slower at those ratios, as Alarmo on my HA takes about an extra 30 seconds or more to May 19, 2020 · Abstract and Figures. The number of cameras does not directly affect the inference speed. With my normal CPU I reach 10ms. If you assumed an average of 3 detections per frame and your 2 cameras are both set to 5 frames per second that equates to (3 * 5) * 2 = 30 inferences per second. The USB Hub (for Power Meter readings) and Coral M2 (for Frigate) are passed through from Truenas to the VM/HA. 13 running in Home Assistant Supervised. Or actually I have added a mean line for the inference speed to my dashboard. It's running on a Ryzen 5600G with 32G of memory and TrueNAS Scale. A single Coral can manage multiple cameras and will be sufficient for the majority of users. 2 Coral, and an external SSD. 2 seconds. But most of the entities that the integration added are "unavailable". host: 192. The samples contain quite a bit of motion and lighting changes. Exports from Frigate API: Inference Speed. Unless it's running at a lower clock or something, but I thought that could only be set in software (drivers installed during the frigate container install). 1% 2. So i have to tinker to find why my inference is ~40 ms with the coral on 4 cams with 5 fps detection framerate. The name frigate in the 17th to early 18th centuries was given to any full-rigged ship built for speed and manoeuvrability, intended to be used in scouting, escort and patrol roles. 0. I'm using the GPU ffmpeg hardware acceleration but I'm curious if there's any benefit to using it with the tensorrt detector or if I should stick to ov? I have 5 camera and the cpu it's running with is an i7 4790k. Camera audio stats. coral must be doing some processing before sending to the TPU (otherwise I'd expect smaller CPU as the processing occurred on TPU?)? Per frame of the camera feed, with multiple (red) motion boxes, does that still send just 1 image to the detector thread, I wonder if many smaller motion boxes (5-10 or so I think I Aug 21, 2021 · Seems like regardless of motion or objects setting in frigate. image Enabled 2 Reolink 810As. Nov 21, 2021 · No branches or pull requests. The term was May 10, 2023 · I have tried modifying frigate the docker img to use the coral-STD firmware (ie standard, not max firmware) - the issue persists with the only difference being the power draw is reduced and inference speed decreased. 0 is out 🎉 with AI acceleration on CPUs and GPU. Jul 13, 2022 · For Home Assistant Addon installations, the config file needs to be in the root of your Home Assistant config directory (same location as configuration. I have a Coral USB accelerator on pre-order but it's going to be a while yet. Did even manage to get my GPU passthrough to work with LXC, did not work with Debian VM (AMD Ryzer 7000 RENOIR integrated GPU) So this works: root@frigate:~# ls -la /dev/bus/usb/002/006. I run a Google Coral and I wouldn't go back to GPU. I've recently moved my Frigate instance to an old PC that I have around which has a GTX 970. I have 11 cameras (7 ip cams and 4 wyze cams). But I'm pretty happy with inference speed around 25ms. Regarding cameras per TPU, from the Frigate website. I saw in release note that this entities would be renamed, but I don't have these either. The Coral USB is plugged into the USB 3 port. 168. Guess that's where the overhead comes from. 5ms, but detector still skips frames pretty regularly. 6 seconds (in ideal conditions) to get a detection. 80 is that correct? (Yes, I've updated the integration as well to 1. I am running Ubuntu 20. 13. Feb 26, 2023 · Frigate comes bundled with, I guess, the drivers for Coral USB because it detects the coral and changes the vendor ID once you run frigate. 9. Resolution and number of cameras is irrelevant to the inference speed. Inference speed is reported around 8. Aug 26, 2021 · Frigate is really designed around having very low inference times so that it's possible to run detection several times on a single frame and still keep up to be realtime. 04. Upcoming Intel CPUs are said to have AI accelerators that improve the inference speeds as well. My inference speed ranges between 6. 27, which lines up with the same speed that is achieved with bare-metal unraid with docker referenced in the Recommended Hardware documentation. 95% and even moving cars have the same score Scrolling through the event list I see older events (created before RC1) and they all have the same score. GET /api/<camera_name> An mjpeg stream for debugging. In today's video, I walked through setting up Axzez's Interceptor 1U case with a Raspberry Pi as a Frigate NVR, or Network Video Recorder. Aug 14, 2022 · My inference speed went from 10 to 50, but my CPU seems to be about the same (1. 3 cameras, low res streams, 5 fps. By far the most informative answer on inference speed, and the advantages of Coral sticks. 0-CBF26E0. Include your full config file wrapped in triple back ticks. However, with my new one the Frigate debug stats are showing 68-78 inference times with the exact same usb coral. Using the latest 0. It has intel HD graphics on-board, same as my old server. Keep in mind the mjpeg endpoint is for debugging only and will put additional load on the system when in use. Then I have a Debian VM that hosts Home assistant Supervised. I only briefly tested with a USB coral, passing through the device (not USB port or hub). Just updated to latest RC1, and as time passed by I noticed that objects in the event list have the same score. 04) with google coral and ffmpeg hardware acceleration - frigate (gitmemory. I have 4 cameras, two 1080p and two 720p. Running on VirtualBo Jan 27, 2023 · hello @aeozyalcin, I am trying to get J5005/openvino working with frigate and struggling to get inference speed lower than 148ms for the bundled ssdlite_mobilenet_v2. I'm running Frigate 0. 0:8443 Is the inference speed ~100ms for 1 cam and ~120ms for 5 camera normal . 2. 859 INF [srtp] listen addr=0. I noticed a Frigate Application in the incubator channel of TrueCharts, but I couldnt get that to work - it bombed out on initialization/config. ADMIN MOD. 12 running with Coral TPU Inference Speed at 7-8ms, so we seem good there. provide examples of inference speeds in the recommended hardware docs page. com) May 2, 2021 · Generally its not efficient to run Frigate inside VM when using Coral as it cause the inference speed to decrease to around 100ms from 6-8ms when running bare metal. yml, it shows the camera when there's any motion detected. frigate controls this speed and it is already set to max. 4-26AE608. And it worked - it did the job and recorded stuff and it was fairly OK at motion detection, but damn did it eat Or are there any other tricks I could try to reduce the peak inference time? Currently I have two CPU processor config's setup with 4 threads each being that the Pi 4 has a 4 core processor. Last refreshed: just now System. 1. The detection fps is the actual number of times per second that the Coral runs object detection. blakeblackshear closed this as completed on Nov 18, 2021. Oct 25, 2021 · Saved searches Use saved searches to filter your results more quickly Jul 4, 2021 · I have been trying to follow the below article to get the google coral recognized with ESXI 7 on a Dell Optiplex 9020 Micro (USFF) but to no avail. Edit: Additional info: My inference speed remains in the same range and does not goes out of it even if i only use one camera. Aug 6, 2022 · Migrating frigate docker from Proxmox VM to LXC caused inference speed went down from 15 to 8ms. Frigate CCTV is absolutely amazing. 2 - one 2280 and one 2230, cpu stays the same in both cases ~15% utilization on i3-9100(4cores). Version of frigate Version 1. Any ideas why tpu inference goes up? A place to share, discuss, discover, assist with, gain assistance for, and critique self-hosted alternatives to our favorite web apps, web services, and online tools. (-) Yolo8n-416: ~19ms. yaml will be MDLefevere (Maurice) July 13, 2022, 1:35pm 3. Similarly, number of cameras is just a function of how many inferences can be run per second. I can confirm frigate run on ESX 7. 4-26AE608 Frigate config file mqtt: host: 192. Many thanks, Laurence. Promox - Home Assistant as VM and there the Frigate addon. At 5 frames per second that is 1 frame every 0. Version. Feb 23, 2024 · Based on CPU, frigate. So 0. 75% of 4 VCPU). Both corals are M. I know VirtualBox isn't recommended (and neither is Reolink) with Frigate, but still thought I should be getting faster inference times. 1 in a docker container on a dedicated Minisforum GK41, with 8 GB of memory, running Ubuntu Server 22. jtbnz. Does frigate "balance" it's work effort between the TPU's or is there some investigating I need to do with the first one as I believe it should be lower No Man's Sky Frigate Calculator. 0. A number of other apps are running on the system, but nothing demanding. Getting inference speeds of 10ms on average with 640x480, at 5FPS, 7 cameras in total. This might be the reason why inference is slow (USB bus lag). Inference optimization is done to improve the performance and efficiency of machine learning models during the inference stage. 2 - so I think that's pretty good given the additional VM and docker. Version 0. 0 and I do indeed get a LED lit with a frigate docker. Jun 19, 2024 · Describe the problem you are having Trying to fix ~150ms inference time. frigate. Frigate config file Aug 6, 2021 · I’m also using HA Blue with USB Coral. A web server is available on port 5000 with the following endpoints. Frigate HassOS aad-on with Coral USB - 20ms+ inference speed (NUC6i3SYK) Support. When using multiple detectors they will run in dedicated processes, but pull from a common queue of detection requests from across all cameras. I am just about to change back to using the iGPU. Oct 21, 2020 · Fyi I've got a usb coral plugged into the back of my HP server which I'm reasonably sure is USB2. With an inference speed of 100 ms your CPU can do 10 inferences in a second… A frigate (/ ˈ f r ɪ ɡ ə t /) is a type of warship. 2 cd ~/Documents/frigate. Once that's set up, you just need to connect Home Assistant and Frigate to that MQTT server you set up. Video Files are stored on dedicated drives (spinning metal, not in the VM space as its on a SSD). May 12, 2021 · Add the Coral drivers (see Coral docs on how to install those and load the modules). yaml or frigate. Testsystem 1: Hyper-V VM running on a Ryzen 3900X, 10 Cores passed to the VM, 8GB RAM. Bad traits will be removed and replaced by beneficial traits while leveling. I've allowed 3 cores as detectors - Frigate logs says to not do this long term. The biggest gap right now in my mind is Frigate's poor feature set for continuous recording, which seems like very basic functionality but ends up as a low priority for a lot of these "event-first" products that are more patterned off of consumer products. With inference times at 150ms, frigate will start to skip frames and/or give up trying to refine bounding boxes. The updated dashboard is here. This PC is more than adequate for Frigate Dec 30, 2021 · Somehow I don't understand the inference speed representations under Debug. 0 and having the models stay loaded in memory I get roughly 10-30ms inference times on usb TPU. I have a less than 6 millisecond inference rate. Nov 22, 2023 · My second coral TPU arrived today and I plugged it into Synology DS1621+ with Docker and I noticed I'm getting mismatched inference speeds between the two TPU's. Now when I infer the same thing in raspberry pi 5 (A76, 8gb ram) the inference speed is just 220 ms per image. (measured by multimeter on usb power or a current meter) Frigate seems like one of the most promising new NVR/VMS products out there, but still lacks the feature-completeness to replace Blue Iris. (Image size 640, batch size 16) While inferring on my laptop using cpu (and ryzen 5 5600, 16gb ram) I am getting around 20ms per image speed. 859514856 12:44:19. 1-367D724 Frigate config file mqtt: host: 192. I am able to see the frigate stream. Note based on my experience: Top left - BEST (inference speed 10-12ms) Top right- disconnected every few hours or so (due to power deficiency) Bottom- not recognized. 85 ms 47. With an inference speed of 10, your Coral will top out at 1000/10=100, or 100 frames per second. 4ghz (4 core) as it was more power efficient. Is it possible for you to provide some insights and sample config you are using and the inference you are getting with J5005? thanks! I picked 5 recordings and created a config. We would like to show you a description here but the site won’t allow us. The reported Inference Speed is 35-40 ms at best, and the cameras are reporting really high CPU usage (50-100%). Oh ok, thx alot for the quick answer. The system should be powerful enough. Config file We would like to show you a description here but the site won’t allow us. Base stats shows the potential of a frigate. Intel NUC NUC7i3BNK: 8-10ms: Great performance. It is expected to be higher than the camera fps sometimes and lower at other times. I just moved to a 7th gen celeron J1900 2. This version starts from a PyTorch model instead of the ONNX model, upgrades the sample application to use TensorRT 7, and replaces the ResNet-50 classification model with UNet, which is a segmentation model. 189 user: mqttuser password: mysupersecureandsecretpassword detectors : coral1 : type: cpu coral2 : type: cpu coral3 : type: cpu birdseye : enabled Sep 27, 2022 · Frigate often runs many detections per frame especially when there are multiple objects in the frame. 9% very slow speed Version SYSTEM 0. Frigate config file So currently running Frigate on an M1 Mac mini using docker. So if they are there for 0. add yourself as a codeowner in the Nov 12, 2020 · Inference Speed Notes; Atomic Pi: 16ms: Good option for a dedicated low power board with a small number of cameras. Camera, detection and skipped FPS. Nearly all objects always has a score of 76. 146 client_id: frigate stats_interval: 60 use My MacBook Pro with an M1 Pro that I use for contributing to frigate has the same inference speed as my "production" device with a USB coral. (Detector Inference Speed) Frigate in Proxmox LXC with Codeproject-AI and Dual Coral TPU question Further information is requested Bettman66 asked Jul 5, 2024 in Question · Answered Jan 13, 2021 · * feat: image support init * feat: LOAD cmd clean up + storage Engine moved to factory pattern * feat: regex support with rollback * fix: undo import image code * fix: code clean up * bug: add missing data files * test: fix old test cases * feat: drop table clean up * bug: remove rename support for structured data * test: remove unnecessary test cases * fix: validate all files before loading Sep 30, 2022 · After upgrading to frigate 0. Also worth mentioning that frigate 0. yaml). only 2 cams currently at 20c ambient. I don't think it matters which Coral you go with. This is an updated version of How to Speed Up Deep Learning Inference Using TensorRT. 67 user: *** pass Aug 9, 2022 · I have two reolink cameras and one amcrest doorbell. Not quite. 75ms. If there is heavy motion on multiple cameras at once it can reach inference speeds of 150ms for a brief moment but mostly hovers from the 20-50ms. Inference speed is 11-12ms Apr 5, 2023 · put me off! Your post gave me some confidence to give efficientdet_lite3_512_ptq_edgetpu. 11. edgetpu INFO : Att Feb 7, 2022 · Describe the problem you are having the person detected, toggle of events and records etc and other relevant sensors are unavailable. Authors present a Multilayer Perceptron (MLP) artificial neural network (ANN) method for the purpose of estimating a speed of a frigate using a combined diesel-electric and Describe the problem you are having Finally got my hands on a coral USB. My cameras are Wyze with the buggy RTSP firmware, but it works quite well considering. Once rebooted with the host machine stopped, add the PCI device to the host (where 101 is the VM ID in Proxmox) root@proxmox:~# qm set 101 -hostpci0 04:00. detector. I got a Coral USB accelerator yesterday and even if it works well overall, I'm far from 10ms when it The Coral will outperform even the best CPUs and can process 100+ FPS with very little overhead. Hoping to get any expereinces with the usb coral from someone who used different speeds yet. Nov 18, 2021 · The detection fps in your config is setting the camera fps for that cameras detect stream. Leverages multiprocessing heavily with an emphasis on realtime over Jan 27, 2021 · It looks like sensor. 0 and 0. HA Supervised has frigate integrated (as Docker). Nov 30, 2021 · Enabled 2 Reolink 810As. Tight integration with Home Assistant via a custom component. Getting inference speeds ~600. I think I saw you are using Wyzecams. If it is still not recognized, try HA restart, or HA host reboot. Describe the problem you are having There are are no objects being recognised. Detection on all is set at their native resolutions. Home Assistant has an official MQTT integration you can just install and for Frigate, you just need to add the mqtt settings to Frigate's config file. Aug 26, 2021 · Minglarn commented on Aug 26, 2021. 1 my cpu with an inference of ~400 will do a better job of detecting on an 8 camera system. coral seems to be the most power efficient overall and is definitely the most power efficient when it comes to speed per watt. yml, but if both files exist frigate. Detection stream resolution: 1280x720 (no HW decoding support) (-) Yolo8n-320: ~12ms. Storage total, used and free. Is this speed considered acceptable with my setup? Would a Coral TPU provide better inference? Version. I pruned and quantised it using sparse ml and then exported it to onnx format. If your detection fps Feb 27, 2022 · I had very low inference times. 13 given the inference speed of 247ms in my p General frigate false positive tips still apply. 1-2EADA21 in docker with compose ( recommended install method ). Device Temperature (Coral temp) My inference speed is about 450 to 650 on each cpu. I can see "inference speed" and detection fps , but can't access the switches to enable/disable detection and snapshot. Previously I was using an nVidia 1660 and it worked, and still does for my video decoding but the coral was a great buy. you should see following output printed in the terminal: STEP 3: Create frigate Docker file. 0 for customer_component I cant se coral_inference_speed, detection_fps, frigate_status and camera fps. Inference is fast on yolo via GPU too but my main concern is how long it takes to actually receive the notification. 1 participant. All-in is under $100. I can send 20 frames of 720p from 5 cameras and it still keeps a 6ms inference speed. I’m not sure which part you got stuck for LXC. You may want to play around with the camera configs to see how it looks. These are the steps I took to get it up and running, it's too easy: Nov 3, 2021 · When using one coral inference speed is ~7ms, when adding second one the inference speed is ~10ms. I'm running 3 cams at 1080p with 22ms inference speed in Frigate. 8-10. I’m confused. Jan 16, 2024 · Inference optimization. See the screen shots below. Add PCIe Device to VM. Try changing the port to the top left. Those switches are unavailable. My inference speed was as high as 13-14 fps at times. If you connect multiple USB Accelerators through a USB hub, be sure that each USB port can provide at least 500mA when using the reduced operating frequency or 900mA when using the I'm running it currently for my setup with HAOS, Frigate, m. By default, Frigate will use a single CPU detector. Update: After running frigate in PVE host, the inference time reduced to 8. First I gave Blue Iris a try which meant I needed a full Windows VM to run it. I have been using Frigates beta and RC for some time now and few days ago they launched a stable version. Hi there, I forgot to mention that I have 2x USB dongles (ZigBee and Z-Wave) plugged to the same NUC. 14. coral_inference_speed is no longer available in 0. That made me even happier. See full list on docs. Author. Aug 29, 2023 · frigate | 2023-11-19 12:44:19. Designed to minimize resource use and maximize performance by only looking for objects when and where it is necessary. I also don't know if not having a gpu is impacting the situation. Frigate NVR 0. Raspberry Pi 4, USB Coral as detector: 9W (including ~1w of a POE splitter and a tiny fan in the case): CPU 40-50%, inference speed 20-30 ms NUC8i5BEK, Proxmox 7, Frigate in a docker container on a Debian LXC, GPU passthrough, OpenVino: 25-35W, CPU 20-30%, GPU 10%, inference 7-9 ms So, it doesn't require it at all, but I just added a m. 12. Jan 24, 2021 · I have a reasonable size Frigate instance at home: 13 cameras; 2 feeds per camera: Main feed 1920x1080 @ 25 FPS: clips, record, RTMP; Sub feed 704x576 @ 6 FPS: detect; Since 2019 I've been using a Coral USB, but after finally selling my old Hikvision NVR on eBay over Christmas, thought I'd invest in reducing my inference speed. For anyone curious I tried upping the resolution to Jun 11, 2021 · Not saying this is a bug - but I am seeing inference speeds around 39. Doing so allows me to plug multiple PoE security cameras straight into the back of the device, and record their IP video streams to disk (the case has space for up to 3 hard drives or SSDs). 129 user: user password: password came Apr 13, 2023 · I have Frigate 0. For inference using usb 3. Jul 23, 2022 · Describe the problem you are having My Frigate CPU usage hovers around 50% and I need to get this down. This is with a Coral connected over USB 3 to Odroid N2+ Trying to figure out if this is expected or something in need of tweaking. just visible feeds in the Frigate UI. I was using a Reolink before and that was working (with only the substream). Other detectors may require additional configuration as described below. My issue is that processing is very slow even the though inference is 20 on a usb 3. Openvino vs tensorrt. With an inference speed of 10, your Coral will reach 1000/10=100, or 100 frames per second. Struggling to find the correct FFMPEG settings however to take advantage of my GPU passthrough, so I receive the notice on Frigate status page that Hardware Accel is not setup. 2 coral and it really sped up it's inference speed. cj xo oa wi fj rw cb sm qa cz