The paper introduces the *Spatial Transformer*, a learnable module that enables spatial manipulation of data within convolutional neural networks (CNNs). This module allows the network to actively transform feature maps based on the input data, without additional training supervision or modification to the optimization process. The spatial transformer can handle transformations such as translation, scale, rotation, and more complex deformations, leading to state-of-the-art performance on various benchmarks. The module consists of three components: a localisation network, a grid generator, and a sampler. The localisation network predicts transformation parameters, the grid generator creates a sampling grid, and the sampler applies the transformation to the input feature map. The spatial transformer can be integrated into existing CNN architectures and is computationally efficient, making it a powerful tool for tasks requiring spatial invariance and transformation invariance. The paper also discusses related work and provides experimental results demonstrating the effectiveness of the spatial transformer on tasks such as image classification, co-localisation, and fine-grained classification.The paper introduces the *Spatial Transformer*, a learnable module that enables spatial manipulation of data within convolutional neural networks (CNNs). This module allows the network to actively transform feature maps based on the input data, without additional training supervision or modification to the optimization process. The spatial transformer can handle transformations such as translation, scale, rotation, and more complex deformations, leading to state-of-the-art performance on various benchmarks. The module consists of three components: a localisation network, a grid generator, and a sampler. The localisation network predicts transformation parameters, the grid generator creates a sampling grid, and the sampler applies the transformation to the input feature map. The spatial transformer can be integrated into existing CNN architectures and is computationally efficient, making it a powerful tool for tasks requiring spatial invariance and transformation invariance. The paper also discusses related work and provides experimental results demonstrating the effectiveness of the spatial transformer on tasks such as image classification, co-localisation, and fine-grained classification.