New build! (0.2.113)

[edge]

Hey everybody,
Hope all’s well out there! Vizy software release 0.2.113 is up and ready to download.

This is a big’n – included in this release:

• Texting support – Vizy can now send you text updates, including pics and short videos. And you can text Vizy commands/queries.
• Rewritten Birdfeeder app – it has improved detection and classification networks, support for European bird species, and texting support
• New Object Detector app
• New TensorFlow detection/classification networks (EfficientDet and EfficientNet)

Here’s how to update your Vizy’s software.

Enjoy!

The Vizy Team

[plord12]

Many thanks for the new release !

Small typo … the birdfeeder docs mention /home/pi/vizy/apps/birdfeeder/birdfeeder_consts.py … this should be /home/pi/vizy/etc/birdfeeder_consts.py

[edge]

Ah, many thanks

[plord12]

So an early success with a European bird in a book -

But first test outdoors many things are identified with a high percentage as a blackbird -

[edge]

Hi Peter,
The new detection network is a significant improvement over the previous one (MobileNet V2), but it still can make detections that make it look fairly silly.

What happens if you disable Common Blackbird in the species of interest?

We’re going to add the ability to share Google Photos albums with us so we can use the pictures to improve the overall accuracy in a semi-automated way. Is your Vizy set up through Google Photos?

Edward

[plord12]

@edge

We’re going to add the ability to share Google Photos albums with us so we can use the pictures to improve the overall accuracy in a semi-automated way. Is your Vizy set up through Google Photos?

Yes, happy to share of course … although nothing yet useful. I expect it will be the summer time until I can gather significant number of pictures.

[plord12]

@plord12

But first test outdoors many things are identified with a high percentage as a blackbird -

Another “silly” example FYI -

[edge]

Hi Peter,
Please try reducing the sensitivity. It may be the case that you’re using the default sensitivity. Some adjustment may be needed.

Edward

[plord12]

@edge

Please try reducing the sensitivity. It may be the case that you’re using the default sensitivity. Some adjustment may be needed.

Yes, trying that now. I’ve setup a test in my garden ( although not a good time of year for birds ).

A couple more mis-identified pics FYA -

[edge]

Hi Peter,
Thanks – the next version will have the ability to share your birdfeeder album with us so we can use your erroneous detections to improve the network (so no need to post your false detections on this forum). Having pictures from false detections in the training set will help improve the accuracy.

Edward

[plord12]

@edge

Hi Peter,
Please try reducing the sensitivity. It may be the case that you’re using the default sensitivity. Some adjustment may be needed.

Edward

I ran some tests this weekend with vizycam and my old (cheap) camera -

Detection sensitivity was reduced to 20% and with the European config, but still mostly false positives at random times -

Didn’t seem to matter where I directed the camera … when the light changed an intruder was detected.

I did just once see a real bird captured -

But wrong bird identified.

My old camera remains much better and taking pictures of actual birds and avoiding non-bird pictures -

( in this case and others, vizycam didn’t capture anything )

I’m wondering if one of the differences here is motion detection.

Finally, as mentioned before, the vizycam IR doesn’t look to be useful -

For comparson, the old camera IR is much clearer -

Given that there are fewer and fewer birds around now, I’ll take down the test tomorrow and put vizy away. I’ll probabally pickup a better version of the old camera.

[edge]

Hi Peter,
Sorry for the problems. Regarding your IR setup, it looks like your Vizy isn’t close enough to the front glass and you’re getting some internal reflection. It also looks like you probably want a narrower field of view either by moving closer to the subject or using a longer lens (e.g. zoom).

Here are some of the setups that have been used with good accuracy/results:

Of note – the lighting is fairly uniform. Still, I can see that you’re frustrated – again, apologies.

Edward

[edro]

Hello Edward,

Sorry I’m so late to the party!

I’ve updated my Vizy software to the latest release (0.2.114) today. Limited testing so far, but it all seems to work for me. Google photos, email, and spreadsheets test ok and the three main apps are running. Like others, I’ve seen false positives in Birdfeeder but I’m not really trying to use it and so haven’t done any tweaking. MotionScope basics work too. I like the new Object Detector app - I think it will come in handy. Its object detection performance seems pretty good so far.

My question: Are you running the latest Raspberry Pi Bullseye release at Vizy HQ - is it safe to upgrade?

Thanks for all your hard work,
Ed

[edge]

Hi Ed,
We’re using Buster version 10 for development currently. Is there a reason you want to upgrade to Bullseye? (I can think of some reasons, but I’m interested in your thoughts/reasoning.)

I think there’s a good reason to upgrade to Bullseye 64 bit OS because it runs a bit faster for most compute intensive applications (such as CNN inference), but we’re punting on this until we get some more applications released. In theory it should be easy, but we want to do full testing before offering it.

Edward

[edro]

It’s not urgent for me. My main motivation is to stay current with security patches / updates. For what it’s worth, I think you’re right to prioritize releasing additional apps.

When the time comes, I’d be happy to help beta test Vizy with Bullseye.