Hi Everyone.
Can someone please explain to me if it’s possible to automatically annotate an image dataset of over 19,000images.
What i mean by automatically is, that there are images of different orientations and tilted at angles. Is it somehow possible to automatically detect this and rotate it differently for each image to make it straight ?
The ultimate goal is to read numbers from the image using pytesseract and OpenCV.
Well, it depends on the type of data you are dealing with. Posting some images can help us to give a useful answer.
For automatic annotation you need an algorithm that can do the job, and this algorithm is highly specialized in the specific problem. Many datasets are hand annotated, and more often than not that’s usually a huge work.
The chicken-egg problem here: if you already have an algorithm to detect whatever you want to detect, then why bother to build a dataset to train a network to do that. Well, there are a number of reasons to do that, like training a new and more efficient or more robust net.
If you want help on your specific problem, you should explain it in a more detailed way, with code and images examples, and the specific annotation you are looking for.
haha nice one @Alejandro_Silvestri agreed!
Fully automatic annotation for all old and new use cases of AI whether computer vision, nlp, speech, documents, LIDAR, Geospatial imagery etc and what not we yet don’t even know about. There are limitless possibilities to singularity in data annotation space. Till that human in the loop and semi automated annotation rocks!
@Orville_Thomas Correct me if i understood properly, but your goal is to read numbers in the image and not necessarily using pytesseract and OpenCV?