This paper proposes a novel edge detection method called Richer Convolutional Features (RCF), which leverages richer convolutional features from all convolutional layers of a deep neural network to achieve more accurate and efficient edge detection. RCF fully exploits multiscale and multilevel information of objects to perform image-to-image prediction. The method is based on the VGG16 network and achieves state-of-the-art performance on several available datasets. On the BSDS500 benchmark, RCF achieves an ODS F-measure of 0.811 with a speed of 8 FPS, outperforming human perception. A faster version of RCF achieves an ODS F-measure of 0.806 with 30 FPS. The method is also versatile and can be applied to classical image segmentation tasks, achieving high-quality perceptual regions. RCF is designed to combine features from all convolutional layers of a neural network, which allows it to capture more detailed and hierarchical information for edge detection. The method uses a novel loss function that considers the reliability of edge annotations, which helps in training the network more effectively. Additionally, RCF employs a multiscale strategy to improve edge detection performance by using image pyramids during testing. Compared to existing methods like HED, RCF uses richer features from all convolutional layers, which allows it to capture more object or object-part boundaries across a larger range of scales. RCF also introduces a new loss function that treats training examples more appropriately, and it uses a multiscale test strategy to further improve performance. Experiments on the BSDS500 and NYUD datasets show that RCF achieves state-of-the-art performance in edge detection, outperforming other methods in terms of both boundary and region quality. RCF is also effective in image segmentation tasks, achieving competitive results when applied to classical image segmentation. The method is efficient and accurate, making it promising for application in other vision tasks. The proposed RCF architecture can be seen as a development direction of fully convolutional networks, like FCN and HED. The source code is available at https://mmcheng.net/rcfedge/.This paper proposes a novel edge detection method called Richer Convolutional Features (RCF), which leverages richer convolutional features from all convolutional layers of a deep neural network to achieve more accurate and efficient edge detection. RCF fully exploits multiscale and multilevel information of objects to perform image-to-image prediction. The method is based on the VGG16 network and achieves state-of-the-art performance on several available datasets. On the BSDS500 benchmark, RCF achieves an ODS F-measure of 0.811 with a speed of 8 FPS, outperforming human perception. A faster version of RCF achieves an ODS F-measure of 0.806 with 30 FPS. The method is also versatile and can be applied to classical image segmentation tasks, achieving high-quality perceptual regions. RCF is designed to combine features from all convolutional layers of a neural network, which allows it to capture more detailed and hierarchical information for edge detection. The method uses a novel loss function that considers the reliability of edge annotations, which helps in training the network more effectively. Additionally, RCF employs a multiscale strategy to improve edge detection performance by using image pyramids during testing. Compared to existing methods like HED, RCF uses richer features from all convolutional layers, which allows it to capture more object or object-part boundaries across a larger range of scales. RCF also introduces a new loss function that treats training examples more appropriately, and it uses a multiscale test strategy to further improve performance. Experiments on the BSDS500 and NYUD datasets show that RCF achieves state-of-the-art performance in edge detection, outperforming other methods in terms of both boundary and region quality. RCF is also effective in image segmentation tasks, achieving competitive results when applied to classical image segmentation. The method is efficient and accurate, making it promising for application in other vision tasks. The proposed RCF architecture can be seen as a development direction of fully convolutional networks, like FCN and HED. The source code is available at https://mmcheng.net/rcfedge/.