@robmarkcole i’ve added extra memory to my VM and installed Deepstack again. In the container I see in the logs that deepstack is working for both teach and recognize. However, when I check the entity in HA it doesn’t look like its being updated when i run the image processing scan. As soon as I run the command in the states page i go to the entity and it shows that it hasn’t been updated since i rebooted HA. Do other people see the same issue?
@Maaniac the deepstack logs willl show if it has procesed any images. Perhaps the VM ports aren’t exposed correctly?
@robmarkcole you mean expose port 5000 to the internet using port forwarding? Why would it then work for object detection and not face detection?
1 Like
No, in my limited experience with VM you have to explicity map the ports from VM to computer
1 Like
@robmarkcole if that would have been the issue then deepstack object would not work either. I’ve installed both using your command in the github page. Ran image processing scan and person detector updated and picked up the different people in the image file. While face counter still shows 0 and has a last update from a few minutes ago when i restarted my HA.
I’m afraid I am out of suggestins then, you should now try the deepstack slack channel
Hi @robmarkcole this looking interesting, as a user of the facebox component what would you say the advantages/disadvantages are over facebox and visa-versa?
1 Like
For face, the advantages of deepstack are more in the medium term, as we are working with the authors to add more features and get a tight home assistant integration. Facebox is stable but we probably won’t be adding any more features to it.
Another difference is that deepstack also allows object detection, which facebox does not
2 Likes
Hello @masterkenobi , Thanks for the useful feedback. We have made new improvements to the Face API based on these. We will release new updates to DeepStack with the following improvements:
- Better result on the face recognition API, delivering 65% to 75% accuracy and more
- Addition of Face Match to the Face API, where you upload 2 images and obtain the similarity score of the faces.
I will keep you updated on this once it is released…
2 Likes
Hi @apt , in addition to the Face APIs , DeepStack provides robust APIs for
- detecting persons, vehicles and 80 most common everyday objects
- recognize and classify scenes to 365 different environmental setups like bedroom, doctor’s office, reception, car park, toilet, etc.
In the next one week , we will release update to DeepStack with the following added :
- FaceMatch API , where you send in 2 images and get similarity scores for the faces
- Moderation API, for recognizing neutral, art, sexy or nudity in images
By the first week of March, 2019, DeepStack will support creating your own custom recognition APIs to recognize any object you want for home, business or industrial applications by allowing you to upload custom recognition models via REST API.
2 Likes
Yes pretty cool. How would you deal between entering and exiting?
My port 5000 is used, I need to change it: with the below command I have to change it how, both 5000 to 5002 (5002:5002)?
sudo docker run -e VISION-DETECTION=True -v localstorage:/datastore -p 5002:5000 deepquestai/deepstack
1 Like
Only change the first 5000 to 5002. So 5002:5000 This is a port mapping to docker. You are telling docker anything received on 5002 goes to 5000 internally within docker. The container is build to use 5000 integral so you can’t change the second number,unless you build it from the ground up.
Home assistant will thing use 5002 on the ip of your docker host to find Deepstack.
I am finding that docker is really the way to go with just about everthing I do in home automation and related.
3 Likes
Thanks!
Started to play with the component, with the following configuration, but my
image_processing.living_room_person_detection stays at unknown state (I ran only the VISION-DETECTION). Any clue, what shall I look for?
EDIT: it is a problem with my camera, not well reachable from homeassistant. Its a problem was trying to resolve since some HASS version, but has nothing to do with this component, which of course need a working camera
you keep the scan_interval under image_processing, or you take it away?
Is it likely that this is going to be brought in as a native HA component once integration has matured?
1 Like
Hi, how to launch the command automatically after a reboot of the host? And should start before/after HASS is running or it doesn’t matter?
I have it but the value is very high. Maybe once every day.