Debug changes

[ErezYosef]

Changes:

• cv2 read images in BGR format. The model trained in RGB input, so in inference it should be the same color space.

• The indexing of scores should be the same as classification and transformed_anchors.
  Otherwise, the presented score will not match the right prediction box.

• Removed unnecessary lines of code in the model. the finalResult list has no further use in the code. It might be there because of previous code versions.

[ErezYosef]

• CPU compliance fix: Change image.coda().float() to image.float()
  There is a few lines above the necessary code:

            if torch.cuda.is_available():
                image = image.cuda()

Fixes issue mentioned: #200 (comment)