This commit aims at addressing uploads of four-channeled images onto the web application predictor tool. For now, such RGBA images are saved, and passed to neural network models; however this last step sends a bug, as the model expects (im_size, im_size, 3)-shaped images, and gets (im_size, im_size, 4)-shaped images.
As a fix, in case of RGBA images, we store only the first three channels and save the image as a RGB picture on the server. A message is also displayed on the webpage to inform the user.
This commit aims at addressing uploads of four-channeled images onto the web application predictor tool. For now, such RGBA images are saved, and passed to neural network models; however this last step sends a bug, as the model expects
(im_size, im_size, 3)
-shaped images, and gets(im_size, im_size, 4)
-shaped images.As a fix, in case of RGBA images, we store only the first three channels and save the image as a RGB picture on the server. A message is also displayed on the webpage to inform the user.
This PR fixes issue #107 .