The data collection process and image labeling

The problem: I like to collect data (images) for an object detection model and it is not clear to me how the labeling will infect the data cleaning and preparation for the model. Is this special for computer vision models due to the fact that it is not easy to drop a column to remove the label(s), or is it?
I am thinking of the feature engineering part. If I label the images will this have an effect on the different methods you can use on the data in order to get better results/performance from your model.
The question is: where in the process is the labeling done? And will it affect the rest of the process?

Images are unstructured. So, there’s no point in representing them as columns (maybe except when storing them). Object detection should have the bounding box information. Unless you augment the image with transformations (in which case the bounding box should change as well), the position of the bounding box will be the same irrespective of the individual pixel values. Depending on the model you use, please follow the instructions of input format and label your data.

Do check deep learning specialization (course 4) since it dives into object detection.

1 Like