Image Classifier on Noisy Labels
PythonTensorFlowMachine LearningCNNClassificationWeak Labels
Implementation of the Multi-Label Fashion Image Classification with Minimal Human Supervision paper from Inoue et al. on the CIFAR-10 dataset.
We carry out model evaluation and selection for predictive analytics on an imbalanced image data.
We will be dealing with a classification problem, where the training labels are not perfect. This is a common phenomenon in data science. Getting accurate ground true labels can be costly and time-consuming. Sometimes, it is even impossible. The weakly supervised learning is a subject that addresses the issue with imperfect labels. In particular, we are going to train a predictive model where label noises exist.