Lightweight Apparel Classification with Residual and Inverted Residual Block based Architectures

Classification of Apparel and Clothing has been the centerpiece in recommendations made for Fashion and E-commerce. This paper explores the applicability of light Deep Learning based classifiers for fast and accurate category classification of images. We use Residual and Inverted Residual Network Based Convolutional Neural Network models, and demonstrate their ability to generalize well and overcome the problems of overfitting. Extensive evaluation on a large dataset with highly class-imbalanced data suggests that the proposed models are fast, compact, and exceed the performance of state-of-the art models with up to approximately 10 times fewer parameters and 4.5 times the speed.