This paper proposes a pipeline for learning deep feature representations from multiple domains using Convolutional Neural Networks (CNNs) for person re-identification. The key idea is to use Domain Guided Dropout, a method that selectively drops neurons based on their effectiveness across different domains. This approach improves the feature learning process by focusing on neurons that are effective for each domain, leading to more robust and generic feature representations. The method is tested on multiple person re-identification datasets, where it outperforms state-of-the-art methods by large margins. The pipeline involves training a CNN on all domains together, followed by domain-specific fine-tuning with Domain Guided Dropout. The results show that the proposed method significantly improves performance, especially on smaller-scale datasets. The effectiveness of Domain Guided Dropout is validated through extensive experiments, demonstrating its ability to enhance feature learning by regularizing the network for different domains. The method is shown to be effective in learning discriminative features that are robust across multiple domains, making it a valuable approach for person re-identification and other multi-domain learning tasks.This paper proposes a pipeline for learning deep feature representations from multiple domains using Convolutional Neural Networks (CNNs) for person re-identification. The key idea is to use Domain Guided Dropout, a method that selectively drops neurons based on their effectiveness across different domains. This approach improves the feature learning process by focusing on neurons that are effective for each domain, leading to more robust and generic feature representations. The method is tested on multiple person re-identification datasets, where it outperforms state-of-the-art methods by large margins. The pipeline involves training a CNN on all domains together, followed by domain-specific fine-tuning with Domain Guided Dropout. The results show that the proposed method significantly improves performance, especially on smaller-scale datasets. The effectiveness of Domain Guided Dropout is validated through extensive experiments, demonstrating its ability to enhance feature learning by regularizing the network for different domains. The method is shown to be effective in learning discriminative features that are robust across multiple domains, making it a valuable approach for person re-identification and other multi-domain learning tasks.