This paper addresses the challenge of training deep neural networks on noisy and incomplete labels, a common issue in visual object recognition and detection tasks. The authors propose a generic method to handle such labeling issues by augmenting the prediction objective with a notion of perceptual consistency. This approach encourages the model to make consistent predictions given similar inputs, where similarity is defined by the deep network's learned features. The method is evaluated on several datasets, including MNIST handwritten digits, the Toronto Face Database for emotion recognition, and the ILSVRC2014 detection challenge. The results demonstrate that the proposed approach significantly improves robustness to label noise and achieves state-of-the-art performance in emotion recognition and object detection, even with minimal engineering effort. The paper also discusses related work and provides a detailed description of the proposed methods, including probabilistic interpretations and experimental results.This paper addresses the challenge of training deep neural networks on noisy and incomplete labels, a common issue in visual object recognition and detection tasks. The authors propose a generic method to handle such labeling issues by augmenting the prediction objective with a notion of perceptual consistency. This approach encourages the model to make consistent predictions given similar inputs, where similarity is defined by the deep network's learned features. The method is evaluated on several datasets, including MNIST handwritten digits, the Toronto Face Database for emotion recognition, and the ILSVRC2014 detection challenge. The results demonstrate that the proposed approach significantly improves robustness to label noise and achieves state-of-the-art performance in emotion recognition and object detection, even with minimal engineering effort. The paper also discusses related work and provides a detailed description of the proposed methods, including probabilistic interpretations and experimental results.