3- Preprocessing the images.

[m-kashani]

Increase the contrast of the important colors. Balanced the red channel. Normalized the red channel.

Then repeat 1.

To Do:

image to image translation -> adverserial

Gan generator / discriminator.

fine tune -> very first layer tune and train the later ones. ( YOLO ).

Style transfering.

• [ ] -> make reds red.

https://caffe2.ai/docs/tutorial-image-pre-processing.html

• [x] https://medium.com/@hirotoschwert/digging-into-detectron-2-part-2-dd6e8b0526e

To read:

• [x] 0: https://web.stanford.edu/~sujason/ColorBalancing/simplestcb.html

  • Try with photoshop.
  • Try with MATLAB.

• [ ] 1: https://gist.github.com/hobson/e3b8805a558d974d48336e133dfb2bbd

• [ ] 2: http://cvil.eecs.yorku.ca/projects/public_html/wb_emulation/files/0105.pdf

• [ ] 3: https://towardsdatascience.com/a-comprehensive-hands-on-guide-to-transfer-learning-with-real-world-applications-in-deep-learning-212bf3b2f27a

• [ ] 4: https://www.kaggle.com/hmendonca/mask-rcnn-and-coco-transfer-learning-lb-0-155

[m-kashani]

cheshmi .... retina object ro detekt kard ya nakard.