This paper introduces two new modules, deformable convolution and deformable RoI pooling, to enhance the transformation modeling capability of Convolutional Neural Networks (CNNs). These modules are designed to learn spatial sampling offsets from the target tasks, enabling adaptive and dense spatial transformations. The proposed modules can be easily integrated into existing CNN architectures and trained end-to-end using standard backpropagation. Extensive experiments validate the effectiveness of deformable ConvNets in various vision tasks, such as object detection and semantic segmentation, demonstrating significant performance improvements over traditional CNNs. The code for implementing these modules is available at <https://github.com/msraver/Deformable-ConvNets>.This paper introduces two new modules, deformable convolution and deformable RoI pooling, to enhance the transformation modeling capability of Convolutional Neural Networks (CNNs). These modules are designed to learn spatial sampling offsets from the target tasks, enabling adaptive and dense spatial transformations. The proposed modules can be easily integrated into existing CNN architectures and trained end-to-end using standard backpropagation. Extensive experiments validate the effectiveness of deformable ConvNets in various vision tasks, such as object detection and semantic segmentation, demonstrating significant performance improvements over traditional CNNs. The code for implementing these modules is available at <https://github.com/msraver/Deformable-ConvNets>.