Hi guys, while annotating some videos with bounding boxes around my objects of interest, I was wandering… If I have 10 seconds of video, 30 fps per second for a total images of 300, if in all these images there is an object I’m interested in but it is static, should I annotate every 300 images with the same identical bounding box? Or should I annotate less images? If I annotate all the 300 images with the same bb, could it be dangerous for a future training? Thank you all.
Hi @JoeYabuki - thanks for posting!
When labeling video, it’s much better to label 1 frame every second or two from lots of different videos than it is to label 300 essentially identical images from one video. Otherwise you’re going to teach the model that the object is always in the same place.
You want diverse data so that your model can generalize in the wild.