Is your feature request related to a problem? If so, please describe the problem:
The problem with our classifier is that it only ranks images with a range of 4.0-7.0. So, the rankings are not very accurate or detailed. An image that is all black will receive a score in the 4.4-4.6 range. This is something that shouldn't happen.
Describe the solution you'd like:
Ideally, an all black image should receive a score close to 0. That will require this classifier to be retrained to a certain degree.
Describe alternatives you've considered:
There are other classifiers out there that do rank images. The Tensorflow library has some of these classifiers, and we have looked at these. However, the classifier we are currently using does exactly what we want it to do, just not as well as it ideally should. I think the problem isn't the classifier, but it is rather the training. If we could just figure out how to retrain it on the images we have been given, that would help its effectiveness a lot.
Is your feature request related to a problem? If so, please describe the problem: The problem with our classifier is that it only ranks images with a range of 4.0-7.0. So, the rankings are not very accurate or detailed. An image that is all black will receive a score in the 4.4-4.6 range. This is something that shouldn't happen.
Describe the solution you'd like: Ideally, an all black image should receive a score close to 0. That will require this classifier to be retrained to a certain degree.
Describe alternatives you've considered: There are other classifiers out there that do rank images. The Tensorflow library has some of these classifiers, and we have looked at these. However, the classifier we are currently using does exactly what we want it to do, just not as well as it ideally should. I think the problem isn't the classifier, but it is rather the training. If we could just figure out how to retrain it on the images we have been given, that would help its effectiveness a lot.