https://mybinder.org/v2/gh/hyemin-lim/plant_classify/HEAD?urlpath=%2Fvoila%2Frender%2Fp_class.ipynb
Using Binder and Google Colab, I made plant classifier.
first image is predicted as succulent, but it is actually cactus. and second image is predicted as cactus, but it is actually succulent. both images are included in the top loss plot because of lack of distinct feature to distinguish them as cactus or succulent.
third image is predicted as cactus but it is actually classified as succulent according to dataset. this image is too low resolution and the individual object is too small to classify.
forth image is predicted as succulent but it is actually cactus, because of the same reason as the third image.(low res&small obj)
fifth image is predicted as succulent but it is actually cactus. the thorn of cactus is not visible so the model is likely to be confused.
sixth image is predicted as cactus but it is actually tree. tree is distinguished by its branch or trunk but in this picture, they are not quite visible.
seventh image is predicted as cactus but it is actually succulent. object other than succulent is occupying over 50 percent of the frame.
eighth ~ tenth image has top loss because of indescribable reason.
resulting action
Here is an data which is predicted as succulent but actually is an cactus. this is top loss image.
