Closed auriml closed 5 years ago
My gut feeling is the large padding around the image due to the small size of the child. In my training data got from openi, many side view X-rays have larger padding than front view ones.
So Just try to remove the padding and test it again to see how it performs.
Hi, I have forked your project to adapt it to python 3.6.1 and pytorch 0.3.1 and try classifying images in other formats (different sizes and uint16). I have uploaded a sample of the new images in https://github.com/auriml/chestViewSplit and implemented a custom loader. It works as expected as far as I have tested it except for this child chest X-ray chestViewSplit/chest_xray/chest_xray_new/7wa3ao.png . Any hint? Thanks!!