This paper presents a method for particular object retrieval using integral max-pooling of CNN activations. The approach improves upon traditional image search systems by integrating CNN-based features into both initial search and re-ranking stages. It introduces a compact feature vector that encodes multiple image regions without requiring multiple inputs to the network. The method extends integral images to handle max-pooling on convolutional layer activations, enabling efficient localization of matching objects. The resulting bounding box is used for image re-ranking, significantly improving the CNN-based recognition pipeline. The method achieves competitive results on the challenging Oxford5k and Paris6k datasets, outperforming previous CNN-based methods. The approach uses a generalized mean to enable integral images with max-pooling, and employs a simple yet effective query expansion method for re-ranking. The method is complementary and, when combined, produces a system that competes with state-of-the-art re-ranking approaches based on local features. The approach is efficient and effective, achieving high performance on large-scale image retrieval tasks. The method is evaluated on the Oxford and Paris building benchmarks, demonstrating its effectiveness in particular object retrieval. The results show that the proposed method outperforms previous methods based on CNN, while being more efficient in practice. The method is based on CNN activations and uses integral max-pooling to enable efficient localization of objects. The method is evaluated on the Oxford and Paris datasets, showing significant improvements in retrieval performance. The method is compared to state-of-the-art approaches and is shown to outperform them in terms of performance and efficiency. The method is effective in both filtering and re-ranking stages, and is able to handle large-scale image retrieval tasks efficiently. The method is based on CNN activations and uses integral max-pooling to enable efficient localization of objects. The method is evaluated on the Oxford and Paris datasets, showing significant improvements in retrieval performance. The method is compared to state-of-the-art approaches and is shown to outperform them in terms of performance and efficiency. The method is effective in both filtering and re-ranking stages, and is able to handle large-scale image retrieval tasks efficiently.This paper presents a method for particular object retrieval using integral max-pooling of CNN activations. The approach improves upon traditional image search systems by integrating CNN-based features into both initial search and re-ranking stages. It introduces a compact feature vector that encodes multiple image regions without requiring multiple inputs to the network. The method extends integral images to handle max-pooling on convolutional layer activations, enabling efficient localization of matching objects. The resulting bounding box is used for image re-ranking, significantly improving the CNN-based recognition pipeline. The method achieves competitive results on the challenging Oxford5k and Paris6k datasets, outperforming previous CNN-based methods. The approach uses a generalized mean to enable integral images with max-pooling, and employs a simple yet effective query expansion method for re-ranking. The method is complementary and, when combined, produces a system that competes with state-of-the-art re-ranking approaches based on local features. The approach is efficient and effective, achieving high performance on large-scale image retrieval tasks. The method is evaluated on the Oxford and Paris building benchmarks, demonstrating its effectiveness in particular object retrieval. The results show that the proposed method outperforms previous methods based on CNN, while being more efficient in practice. The method is based on CNN activations and uses integral max-pooling to enable efficient localization of objects. The method is evaluated on the Oxford and Paris datasets, showing significant improvements in retrieval performance. The method is compared to state-of-the-art approaches and is shown to outperform them in terms of performance and efficiency. The method is effective in both filtering and re-ranking stages, and is able to handle large-scale image retrieval tasks efficiently. The method is based on CNN activations and uses integral max-pooling to enable efficient localization of objects. The method is evaluated on the Oxford and Paris datasets, showing significant improvements in retrieval performance. The method is compared to state-of-the-art approaches and is shown to outperform them in terms of performance and efficiency. The method is effective in both filtering and re-ranking stages, and is able to handle large-scale image retrieval tasks efficiently.