Image compression optimized for 3D reconstruction by utilizing deep neural networks

Computer vision tasks are often expected to be executed on compressed images. Classical image compression standards like JPEG 2000 widely used. However, they do not account for the specific end-task at hand. Motivated by works recurrent neural network (RNN)-based and three-dimensional (3D) reconstruction, we propose unified architectures solve both jointly. These joint models provide tailored task of 3D reconstruction. Images our proposed models, yield reconstruction performance superior as compared using compression. Our significantly extend range rates which is possible. We also show that this can done highly efficiently almost no additional cost obtain top computation already required performing task.