Hand labeling is a cheap trick to increase a validation score on small subset of data, which limits the effectiveness of model when it's implemented in real world.
@amyflorida626 Can we make correction using on those images that are wrongly labelled based on what our image preprocessing exploratory analysis/method revealed to us.
Hello @amyflorida626. I would like to clearly understand what hand labeling means for this competition. For this competition, the NUMBER OF PALM TREES is the LABEL. Keeping this in mind will the following strategies be considered hand labeling?
Strategy 1: Manually crop a few palm trees from images and use them as a template to recognize palm trees in images.
Strategy 2: Manually draw bounding boxes around palm trees and use this data to train models.
Hello @amyflorida626. Any reason as to why this rule is in place? I am asking because it does not mimic the real-world.
Hand labeling is a cheap trick to increase a validation score on small subset of data, which limits the effectiveness of model when it's implemented in real world.
Yes exactly @theuntitled.
@amyflorida626 Can we make correction using on those images that are wrongly labelled based on what our image preprocessing exploratory analysis/method revealed to us.
Hello @amyflorida626. I would like to clearly understand what hand labeling means for this competition. For this competition, the NUMBER OF PALM TREES is the LABEL. Keeping this in mind will the following strategies be considered hand labeling?
Strategy 1: Manually crop a few palm trees from images and use them as a template to recognize palm trees in images.
Strategy 2: Manually draw bounding boxes around palm trees and use this data to train models.