I’m a deep learning engineer who is happy to provide labeling/annotation feedback to improve Wyze AI models.
Should all present objects in a video be labeled, even if they are not moving? For example, if I walk out to my driveway to pick up mail, I see a motion ROI indicator on me in the recording. For the feedback, it is clear I select (or leave selected) “Person” though with my parked car in the FOV, should I also select (or leave selected)“Vehicle”?
If the wind is blowing branches on a tree or bush that triggers motion detection. do I not label it at all, or do I label it “Nothing”?
It might help to have negative labels such as “Wind effects”, “spider”, and “spider web”, from the extraneous motion hits that I’ve seen.
I’d like to point out that it is extremely important to label the videos for the purposes that they will be trained for. If some people only label moving vehicles and other label any vehicle in the image, the training results will be significantly impacted.
Wyze, please provide unambiguous guidance on how to label the videos. Otherwise, I hesitate to provide this labeling feedback as I don’t know if I’m helping or detracting.
yes, plz label all the things you can see, even if it’s moving.
you can label it as “Nothing”
We’ll add the labels based on your feedback
In general, once you submit the videos, all of them will be used for training. Out in house data annotation team will process the data. By adding the label there can help us accelerate the process.