Face and person detection with Deepstack - local and free!

Share your Projects!
484

Please stick with the one I linked directly

485

Okay. I upgraded, but I will overwrite the file again.
By the way, did the naming change of the files saved?
It adds the name value in front of the deepstack_latest_object / car whatever is chosen.

I might have missed it somewhere, but I just want to make sure I didnā€™t screw anything up here.

Will get back to you when tested.

486

I donā€™t know what version you were on before. All changes are in release notes.
However I am beginning to suspect this is not an issue with the deepstack integration but actually from the camera

487

I try to be on the latest release at all times. Now I run the latest with your modified file you provided me with.
I added the time stamped feature and I do see the files coming in just fine. Every single one of them is ā€˜differentā€™.
When I monitor the cameras in ā€˜real timeā€™ I see the exact same thing being snapped and saved as a deepstack file.

I donā€™t see how this could be a camera problem?
Iā€™ve added a lot of delay, so when the image_processing feature is fired, itā€™s more than 10 seconds after the latest processing.

I need to test it fully today and get back to you.

488

I am using Blink cameras, and sometimes the updated image is actually still the last image

489

Arh okay!

Hmm, but when you say updated. Are you then referring to the image saved in the library?

490

IE the same image is being processed at 2 different times. I am saving all images so can review the history and see that is the case

491

Hey Rob,

Maybe this is just a coincidence, but for the last two days I havenā€™t got a notification with the same picture.
Not even when triggering all 4 cameras at the same time, running around the house like a mad man :slight_smile:

So something seems to have been improved?

492

OK well sounds like the changes I made in that file have worked :slight_smile:

493

@robmarkcole using the latest and works fine as well. Small note for your consideration in future changes. You changed the timestamp in the .py like so ā€œ%Y-%m-%d %H:%M:%Sā€ leaving a gap between date and time.
This provides an incomplete link and some messanger services (Gotify) cant complete and the picture doesnt pull through.
In your future release please add it back if possible.
Thank you.

1 Like
494

@juan11perez and @Yoinkz go get release v2.4 which addresses your respective issues.
Cheers

1 Like
495

Done, thank you very much

496

I really like deepstack, @robmarkcole thanks for the component!
I just want to ask if there is a way to reduce CPU usage in any way: i have a vps and a local nuc without avx and 3 of 4 /app/intelligence.py processes are always using between 8 to 15% CPU.
Other than this i find deepstack very strong as an object detection.

Iā€™m a bit OT but: i wish @robmarkcole could port to HA Azure vision intelligence ( not face ) as another object detection service, since there is a free tier :grinning:

497

Hi,

For some reason I no longer seem to get anything fired on the event bus under image_processing.file_saved when deepstack runs. I do get a file createdā€¦ This happened after one of the latest upgrades. Anyone else see this?

1 Like
498

Figured it out. Without the new(?) ā€œsave_timestamped_file: Trueā€ nothing fires on image_processing.file_saved. I mistakenly thought the overwrite of the last file would suffice.

499

Thanks man! Your hard work is much appreciated!!!

1 Like
500

Is there any reason behind this suggestion, e.g. you have experience with it or it has some unique functionality? Iā€™ve also published custom components for google vision and AWS rekognition, also have a PR open adding sighthound

501

well, i tried a bit azure intelligence and for my test cases itā€™s really really good.
Beside that, google vision has 1000 calls while azure raise up to 5000, like sighthound, but i donā€™t find the last one ( sighthound ) reliable ( often sent me message for a person while it was an hangerā€¦ ).
I didnā€™t tested rekognition BTW, apart from the tier that is not so great and i didnā€™t found if the picture frame is saved or not after detection.
Since you are the object detection master of HA, i just gave you an hint :grin:

502

Hey Rob, I was too fast. I updated the Deepstack integration a few hours ago and now I see the same images being firesd again.

Should I try copying the image_processing file from your previously mentioned post into the library again?

503

Those changes are merged into the latest release. If this issue still persists then I think it is an issue with the camera