This paper proposes a deeply-recursive convolutional network (DRCN) for image super-resolution (SR). The DRCN has a deep recursive layer with up to 16 recursions, which increases the receptive field without adding new parameters. However, training such a network is challenging due to exploding/vanishing gradients. To address this, the authors propose two extensions: recursive-supervision and skip-connection. Recursive-supervision uses the output of each recursion to reconstruct the target image, while skip-connection connects the input to the reconstruction layer, improving performance by preserving input information. The DRCN outperforms previous methods on benchmark datasets. The DRCN consists of three sub-networks: an embedding network, an inference network, and a reconstruction network. The embedding network converts the input image into feature maps, the inference network applies recursive convolutions to expand the receptive field, and the reconstruction network generates the high-resolution image. The inference network uses a single convolutional layer repeated multiple times, which increases the receptive field without increasing the number of parameters. The reconstruction network then maps the feature maps back to the original image space. The authors evaluate the DRCN on several benchmark datasets, including Set5, Set14, B100, and Urban100. The results show that the DRCN achieves higher PSNR and SSIM values compared to existing methods. The DRCN also performs well in preserving sharp edges and details in the super-resolved images. The method is trained using mini-batch gradient descent with a learning rate of 0.01 and a weight decay of 0.0001. The model uses 16 recursions, which increases the receptive field to 41x41, and achieves state-of-the-art performance on the benchmark datasets. The DRCN is also compared with other state-of-the-art methods, and it outperforms them in terms of both quantitative and qualitative results. The authors conclude that the DRCN is a promising approach for image super-resolution, and further research can explore deeper recursions to utilize more image-level context.This paper proposes a deeply-recursive convolutional network (DRCN) for image super-resolution (SR). The DRCN has a deep recursive layer with up to 16 recursions, which increases the receptive field without adding new parameters. However, training such a network is challenging due to exploding/vanishing gradients. To address this, the authors propose two extensions: recursive-supervision and skip-connection. Recursive-supervision uses the output of each recursion to reconstruct the target image, while skip-connection connects the input to the reconstruction layer, improving performance by preserving input information. The DRCN outperforms previous methods on benchmark datasets. The DRCN consists of three sub-networks: an embedding network, an inference network, and a reconstruction network. The embedding network converts the input image into feature maps, the inference network applies recursive convolutions to expand the receptive field, and the reconstruction network generates the high-resolution image. The inference network uses a single convolutional layer repeated multiple times, which increases the receptive field without increasing the number of parameters. The reconstruction network then maps the feature maps back to the original image space. The authors evaluate the DRCN on several benchmark datasets, including Set5, Set14, B100, and Urban100. The results show that the DRCN achieves higher PSNR and SSIM values compared to existing methods. The DRCN also performs well in preserving sharp edges and details in the super-resolved images. The method is trained using mini-batch gradient descent with a learning rate of 0.01 and a weight decay of 0.0001. The model uses 16 recursions, which increases the receptive field to 41x41, and achieves state-of-the-art performance on the benchmark datasets. The DRCN is also compared with other state-of-the-art methods, and it outperforms them in terms of both quantitative and qualitative results. The authors conclude that the DRCN is a promising approach for image super-resolution, and further research can explore deeper recursions to utilize more image-level context.