<p>Recent advanced deep-learning studies have shown the positive role of feedback and multi-scale mechanism in image super-resolution (SR) tasks. However, both of them are still in the early stages of excavation, there are more possibilities to explore. Therefore in this paper, to explore the potential of feedback and multi-scale mechanism we design a novel network structure that consists of both feedback mechanism and multi-scale structure (we call it multi-scale feedback block). Multi-scale structure allows our model to exploit the image features fully, and the feedback mechanism enables the model to generate powerful high-level representations. By reasonably stacking multi-scale feedback block we construct a multi-scale feedback residual network for single image super-resolution (SISR). Besides that, we also introduce curriculum learning mechanisms into our network to accelerate and stabilize training. Our detailed experimental results strongly demonstrate that our proposed model is effective and practical.</p>