0

I was wondering if having images with watermarks in the training set when training a CNN (MobileNetv2) will affect its accuracy. Would simply removing them from the set be a better thing to do? Thank you!

Russell C.
  • 588
  • 6
  • 25
  • how much % of the images you have with watermark ? do you expect your real time (during deployment of the model) images to have watermark ? – venkata krishnan Jul 25 '19 at 03:51
  • About 5%-10% of the images in the training set have a watermark. I do not expect my real time images to have a watermark. @venkatakrishnan – Russell C. Jul 25 '19 at 04:47
  • then probably you can remove the images with watermark. if the watermarks are on the edges alone, try cropping it. If you feel the data is less, you can always try augmentations techniques. Let me know the results, without the watermark images. – venkata krishnan Jul 25 '19 at 05:21

0 Answers0