Q: What is Smart Vision? How’s Smart Vision different from the regular Cam Protect Smart Detection?
A: In the past, to build the AI to recognize an object on a webcam, you need to train an AI model first. Think of it as the process of teaching a kid to learn. For almost all AI object detection technology in the market, the approach is to find the patterns among millions of training materials and build a model that fits most situations. This approach comes with a disadvantage in nature - it cannot fit into every single personalized use case. With Wyze Smart Vision, Wyze users are provided with an easy-to-use system that they can build and train their own AI model for the specific use case of the Wyze Cam. For example, you can train a model to detect if the door is opened or closed.
Q: Is Smart Vision compatible with Cam Protect smart detection?
A: The Smart Vision is exclusive to the Cam Protect smart detections (Person, Package, Vehicle, Pet, etc.) If you trained a Smart Vision model on your device, you’ll only get the Smart Vision detection result. If you would like to change back to Cam Protect smart detection, you can do this by simply turning off the Smart Vision model on the skill page.
Q: Who has access to Smart Vision?
A: Right now the Smart Vision project is open to test for Cam Protect users first. You’ll need an active Cam Protect license bound to the device to be able to test.
Q: What devices are supported by Smart Vision?
A: Smart Vision supports Wyze Cam v2, Wyze Cam v3, and Wyze Cam OG for now. Support for more devices is coming later.
Q: What is a Smart Vision skill?
A: Just like teaching a toddler to speak and walk, with Smart Vision, you can teach Wyze cameras skills to detect specific things that you care for. For example, the skill can be to detect if a stovetop is left on, or if a door is not locked properly. Once the skill is learned and equipped, the camera will be able to perform the skill and improve it as time goes by.
Q: Can I create multiple Smart Vision skills for one camera?
A: Yes. Multiple skills can be created for each Smart Vision camera.
Q: How to set the area of interest properly?
A: In the skill set-up process, you can set your Attention Zone by simply dragging the green box to where you want the camera to focus. The tip is to make sure that within the Attention Zone, you can observe the difference between your labels clearly - so the camera can learn the difference easier later on.
Q: Why can’t a user change the detection zone once it has been set up?
A: Since Smart Vision learns the differences between different sets of images that are grouped under labels, changing the detection zone will essentially create too much noise and confuse the model. If the previous detection zone doesn’t work for you anymore, we recommend just creating a new skill.
Q: Why my skill is not accurate?
A: We are sorry to hear that! The skill accuracy is dependent on three main things: data quality, data quantity, and how complex the job is. Training cameras to recognize apples from bananas in a clean contrast background can be fairly easy while recognizing apples from peaches in a noisy, crowded background can be much harder.
That being said, we recommend you consider these questions when setting up a new skill: How clear a difference there is between my labels? Is my Attention Zone set up properly to focus on the object I want to detect and potentially exclude distractions? Are the images I uploaded good quality? Am I feeding each label enough images?
While you are feeding better data to the camera to help it learn, our team will strive to improve the model so it can provide better performances more efficiently.
Q: How long does it take to train a skill?
A: We know it’s disappointing but, it depends. Training could be done in as quick as milliseconds, or as long as a few hours. Training speeds depend on how crowded the server is (how many other users are also training at the same time) and how many images in your skill need to be trained. If there’s too much traffic in the future, we will consider adding servers to ensure a smooth experience.
Q: When do I need to re-train a skill?
A: You can re-train a skill at any time you want! If you find the model is not accurate enough, feed it with more/better images to let it learn!
Q: How do I set up notifications for my Smart Vision labels?
A: You set it up when creating skills. In the current version, you can only create two labels under each skill, one being the default state and the other one being the of-interest state. For example, if you are setting up your skill to detect if you leave your door open, then the door closed will be the default label, and the door opened will be the of-interest label and our Wyze camera will send you notifications whenever it detects the door opened.
Q: Can I create more than two labels for one skill?
A: Currently we only provide a binary model meaning that you can only create two labels under each skill. Support for more labels is coming later - stay tuned!