@wittaj Thanks. I'll clone the camera in the AM and set it up to look for amazon, dhl, fedx and ups. I'm hoping they'll be big enough in the view to get recognized and let you know what happens later tomorrow.
Sweet! I got it to work once in the tuning option in playback, but most of the time several of us here have the rectangle box blip on and disappear really quick, so some sort of bug.
FedEx I think will have the best shot as their logo is the largest of them.
I'll try the ones I mentioned and verizon, too. My neighbor works for Verizon and has a van that goes by at least twice a day during daylight hours so that will be a test as well.
It's been a little crazy here since I said I'd give it a go. I have added the logo.pt file and added some logos to the cameras detection but haven't had the time to go back and look at things yet. I'll get to it a little later and post my results here. I did do a quick look-see this morning and I don't have a lot of hope, so far.
I sent an email to Ken with these proposals to help when we’re debugging DeepStack performance... (no response yet)
1. Would it be possible to make the 'Testing & Tuning > Analyze with DeepStack' option reflect the entries in the 'Custom models' box for the active profile?
2. And/or could the 'Testing & Tuning > Analyze with DeepStack' option display a popup with the same info as the &JSON macro for alerts?
Quoting myself to add the following responses from Ken...
1. Using camera's current settings for objects with 5.4.7.11. (6/11)
Note: the emphasis is mine; I've asked if the camera's current settings will also work for faces & custom models.
2. I like the idea generally about displaying the raw JSON simultaneously. I would need to work on that as a feature suggestion thanks. (6/8)
OK. I just checked and got a logo capture, at least it captures the logo using "tuning". It did identify the FedX logo very early in the clip and also identified that it was a truck. The clip is actually listed as a truck and not FedX which is a little disconcerting but I guess that's because the confidence level is higher for the truck. I didn't notice any blue border, either.
Quoting myself to add the following responses from Ken...
1. Using camera's current settings for objects with 5.4.7.11. (6/11)
Note: the emphasis is mine; I've asked if the camera's current settings will also work for faces & custom models.
2. I like the idea generally about displaying the raw JSON simultaneously. I would need to work on that as a feature suggestion thanks. (6/8)
The Testing & Tuning > Analyze with Plate Recognizer option currently displays the json response in a pop-up. A bit rudimentary but it works. Hopefully, your request for Deepstack json output isn't too far away.
OK. I just checked and got a logo capture, at least it captures the logo using "tuning". It did identify the FedX logo very early in the clip and also identified that it was a truck. The clip is actually listed as a truck and not FedX which is a little disconcerting but I guess that's because the confidence level is higher for the truck. I didn't notice any blue border, either.
That is exactly what it will do. The truck is from the DS model that does not have fedex in it. If DS had fedex native in it, we wouldn't need the openlogo custom model.
Did it capture it in the live view or just playback?
I believe playback will show all of the items DeepStack can identify? At least it always has with mine LOL. I am getting potted plants, toothbrush, LOL.
I believe playback will show all of the items DeepStack can identify? At least it always has with mine LOL. I am getting potted plants, toothbrush, LOL.
That has been my experience as well. Not sure where there is a chair on my stairs, but yeah, it seems to tag anything it finds in analyzing. No complaints though.
@wittaj The only time FedX showed up was during a playback of the clip with tuning on. Watching that clip I also found out I have a subway right in my own front yard! I think that means the train, not the sandwich. There's a rock garden full of irises in the full frame and DS sees it as a "subway". I also had to move the dark.pt file out of the models folder to get it to work. Now I'm wondering if DS can only handle one additional model at a time. Last night I tried using subdirectories, one for dark and one for logos, but that was a fail. Apparently BI only looks in the directory on the "AI" config page and won't look in a subdirectory, under the main models directory, specified in the camera configuration. If that could happen it would solve lots of problems and cut CPU spikes down both in time and intensity. The dak model hasn't really helped me at all so I'll leave the logo model in use and see how it works over a longer time period.
Anyone know if there's a limit to the number of objects that can be specified for detection?
I believe playback will show all of the items DeepStack can identify? At least it always has with mine LOL. I am getting potted plants, toothbrush, LOL.
The images for several of my cams seem to be target-rich environments, because I keep getting 6-10 overlapping rectangles and labels. Often this yields a less-than-helpful marked-up image because the tag I’m trying to verify per my settings is obscurred a “potted plan” or a “boat” or … I like seeing what’s going on under-the-hood, but would prefer “seeing” it with the &JSON output.
And, just how in the the world can my neightbor’s Great Dane always be tagged as a bicycle? LOL
I have sit back for the most part and watched this evolve. I finally took the leap yesterday and installed Deepstack and set it up, on 4 cameras.
Results have been very mixed, I can only get it to work on two of the four cameras reliably. And by work, I mean deepstack actually working to record a clip.
The two cameras I have issue's with, both show deepstack doing it's job when using test and fine tune, but for the most part, deepstack doesn't do anything when just letting it do it's thing.
I have sit back for the most part and watched this evolve. I finally took the leap yesterday and installed Deepstack and set it up, on 4 cameras.
Results have been very mixed, I can only get it to work on two of the four cameras reliably. And by work, I mean deepstack actually working to record a clip.
The two cameras I have issue's with, both show deepstack doing it's job when using test and fine tune, but for the most part, deepstack doesn't do anything when just letting it do it's thing.
I use it for most but a couple it's just not practical.. this camera here for instance is probably best flawless with alerts and sending MMS trigger pictures. Memo might not always be accurate but I never miss a person or vehicle coming in and it's quicker to see a text than open bi .. I'm extremely happy with this particular setup
I had a similar experience when I first started. Obvious things wouldn't trigger in DS. Now it seems to work fairly well although it still misses an occasional obvious target. I guess it's in the 95% range in accuracy. It's running on most of my cameras so it's getting older 2MP and newer 4MP video which means I can't pin the fails on the cameras.
One other comment is that the target size and contrast seem to be on the critical side. Example being a light colored car in bright sunlight can be missed and a dark color car in late evening can be missed.
What version of BI are you running? I'm on the "latest and greatest" 5.4.7.11
Note; I am tired, had to edit out the typos three times, so far!
github.com
