This paper addresses the challenge of learning generic and robust feature representations from multiple domains for person re-identification. The authors propose a pipeline that combines Convolutional Neural Networks (CNNs) with Domain Guided Dropout (DGD) to improve feature learning. The pipeline first trains a CNN jointly on multiple domains using a single softmax loss, then identifies the effectiveness of each neuron on each domain by computing impact scores. DGD is applied to dropout neurons based on these scores, guiding the CNN to learn more discriminative features. The method is evaluated on several person re-identification datasets, showing significant improvements over state-of-the-art methods, with the largest gain of 46% on the PRID dataset. The paper also discusses the effectiveness of different DGD schemes and their impact on the network's performance.This paper addresses the challenge of learning generic and robust feature representations from multiple domains for person re-identification. The authors propose a pipeline that combines Convolutional Neural Networks (CNNs) with Domain Guided Dropout (DGD) to improve feature learning. The pipeline first trains a CNN jointly on multiple domains using a single softmax loss, then identifies the effectiveness of each neuron on each domain by computing impact scores. DGD is applied to dropout neurons based on these scores, guiding the CNN to learn more discriminative features. The method is evaluated on several person re-identification datasets, showing significant improvements over state-of-the-art methods, with the largest gain of 46% on the PRID dataset. The paper also discusses the effectiveness of different DGD schemes and their impact on the network's performance.