This paper proposes a part-based R-CNN method for fine-grained category detection, which overcomes the limitations of previous methods by leveraging deep convolutional features computed on bottom-up region proposals. The method learns whole-object and part detectors, enforces geometric constraints between them, and predicts a fine-grained category from a pose-normalized representation. Experiments on the Caltech-UCSD bird dataset show that the method outperforms state-of-the-art fine-grained categorization methods in an end-to-end evaluation without requiring a bounding box at test time. The method uses deep convolutional features to detect both whole objects and their parts, and applies non-parametric geometric constraints to improve part localization. The system is trained on a dataset with ground truth bounding boxes and part annotations, and is tested on the Caltech-UCSD bird dataset, which contains 11,788 images of 200 bird species. The method achieves high accuracy in both scenarios where the ground truth bounding box is known and where it is not. The method uses a combination of deep convolutional features and geometric constraints to improve part localization and fine-grained categorization. It is evaluated on the Caltech-UCSD bird dataset and shows superior performance compared to previous methods. The method is also tested on a scenario where the ground truth bounding box is not known, and still achieves high accuracy. The results show that the method is effective in both scenarios and can be used for fine-grained categorization without requiring a bounding box at test time. The method is also compared to other state-of-the-art methods and shows superior performance. The method is also evaluated on the Caltech-UCSD bird dataset and shows superior performance compared to previous methods. The method is also tested on a scenario where the ground truth bounding box is not known, and still achieves high accuracy. The results show that the method is effective in both scenarios and can be used for fine-grained categorization without requiring a bounding box at test time.This paper proposes a part-based R-CNN method for fine-grained category detection, which overcomes the limitations of previous methods by leveraging deep convolutional features computed on bottom-up region proposals. The method learns whole-object and part detectors, enforces geometric constraints between them, and predicts a fine-grained category from a pose-normalized representation. Experiments on the Caltech-UCSD bird dataset show that the method outperforms state-of-the-art fine-grained categorization methods in an end-to-end evaluation without requiring a bounding box at test time. The method uses deep convolutional features to detect both whole objects and their parts, and applies non-parametric geometric constraints to improve part localization. The system is trained on a dataset with ground truth bounding boxes and part annotations, and is tested on the Caltech-UCSD bird dataset, which contains 11,788 images of 200 bird species. The method achieves high accuracy in both scenarios where the ground truth bounding box is known and where it is not. The method uses a combination of deep convolutional features and geometric constraints to improve part localization and fine-grained categorization. It is evaluated on the Caltech-UCSD bird dataset and shows superior performance compared to previous methods. The method is also tested on a scenario where the ground truth bounding box is not known, and still achieves high accuracy. The results show that the method is effective in both scenarios and can be used for fine-grained categorization without requiring a bounding box at test time. The method is also compared to other state-of-the-art methods and shows superior performance. The method is also evaluated on the Caltech-UCSD bird dataset and shows superior performance compared to previous methods. The method is also tested on a scenario where the ground truth bounding box is not known, and still achieves high accuracy. The results show that the method is effective in both scenarios and can be used for fine-grained categorization without requiring a bounding box at test time.