He, Kaiming, et al. "Spatial pyramid pooling in deep convolutional networks for visual recognition." IEEE transactions on pattern analysis and machine intelligence 37.9
implementaion of SPP-net paper using PyTorch. In this project a mutliscale and multilabel classifier is trained and evaluated Using Voc pascal dataset.
Network is trained with N epochs, such that each epoch is performed with different scale.
binary cross entropy (with logits loss) is used with weight of positive propotional to the probability of the class exitance for i.e) if the positive class ocurred k times of a dataset of size N the weight of positive loss is k/N.