Experimental Setup

Implementation : GitHub

We trained all the models using GPU on Google Cloud Platform using a Deep Learning VM instance with the following configurations -

• VM instance: 2 vCPUs + 13 GB memory (n1-highmem-2)

• Standard Persistent Disk: 100GB

• NVIDIA Tesla K80 GPU

• TensorFlow Enterprise 2.1 (CUDA 10.1)

• Salt and Pepper Noise

• Low Resolution (LR) images generated by downscaling and upscaling by a scale factor and using nearest neighbor interpolation for the upscaling task.

• Dataset - PASCAL VOC 2007

MemNet[1] 6 memory blocks and 6 recursive units per memory block is the model architecture used for all the experiments.