It’s a J3455 Celeron, what do you think?
Yeah, 10 seconds seems a little long but doesn’t seem totally out of the question… Usually the first detection will take a long time and then subsequent ones will be faster…
Did you think that a little bit more of ram can be usefull? Or is totally not related to ram?
Decreasing the resolution can help speeding up the process?
I’ve tried tflite detector, super fast but a little bit unrealible… I need to detect cat in my garden and open the sprinklers since they use it like a toilet 
So 10 seconds is too much to work, there is some faster tensorflow model useful for me then the integrated one in the docker?
Sadly more memory won’t likely help. It’s not likely related to ram unless it’s swapping. It would likely take longer than 10 seconds though. Honestly, the only thing I can recommend is trying to get a different computer. I use old HP EliteDesk desktops. They have a Core i5-6600 and it can do a detection in about a second. They can be had on ebay pretty regularly for less than 100$ if you shop around. I got one with 16GB and slapped a cheap SSD in it for about 100$ all in.
This can be a good idea, i saw a bunch of renewed i5 small desktop machine on amazon for about 100€…
There’s any comparison about detection time for cpu somewhere?
I haven’t seen any… but I can tell you my i5-6600 with 16GB does a tensorflow detection in just over 300ms. It does a pytorch yolo detection in about 70ms…
i’ve tried with a lower resolution (from 1080p to 720p), the second detection took about 3s, something usable…
But i’ve some problem detecting cats, this night a spider on the lens appear as a cat witha confidence of 98% and with this 2 images the detector can’t find nothing. What do you suggest for animal detection? Tensorflow or yolo? Another model instead of the one preinstalled in your container?
I really don’t have any suggestions… It may be that your camera is too close. Other than that, you may need to just try different models. yolo may work better. It’s really hard to say other than maybe training a custom model which I still haven’t figured out how to do.
Honestly, an infrared detector in your case is likely to be a lot more reliable.
I’m really dumb with linux, i’ve tried using the guide on github to use the yolo model but i don’t have a clue where to start. There’s an “idiot” step by step tutorial somewhere?
Yolo is faster then tensorflow?
Last questiont, how can i see the log of the detection using the docker version?
You should just be able to use Yolo. It’s part of the default config. The easiest thing to use is docker logs -f <container name> to see the logs.
Hello. Is it possible to specify an object detection zone?
There’s gobs of options including detection areas… GitHub - snowzach/doods2: API for detecting objects in images and video streams using Tensorflow
1 Like
Visually it will not work? As in Frigate
Not sure what you mean?
Nope it doesn’t do that… It’s not really an interactive NVR like frigate. It’s just a simple API.
Do you have an example of how to create a zone for detecting objects?
If i get ‘DetectResponse(id=’’, image=None, detections=[], error=None, duration=553.1625429999849)’
means that nothing is detected? Or can be something detected but below the confidence set?
Nothing was detected for the parameters set. You can lower the confidence. It almost always detects something but it’s not right unless the confidence is high enough.