Please notice that frames were motion compensated and were normalized by dividing 65535 before feeding into VBDE network. Only VBDE net was provided in our work. You can use your own MC algorithm to get motion compensated frames.
Set your own configuration like caffe path, model path, pretrained parameter path and result path in test_VBDEnet_4to16.m. Please guarantee video frames were motion compensated before feeding into VBDE network.
For ‘Sintel’ test images, run test_VBDEnet_4to16.m to get 16-bit image. We provide four patches in MAT formats of one image that was motion compensated in test_data directory, you can have a try directly with provided data without pre-processing.
If you use this code, please cite the following publication: J.Liu, P.Liu, Y.Su, P.Jing, X.Yang, "Spatiotemporal Symmetric Convolutional Neural Network for Video Bit-Depth Enhancement", to appear in IEEE TRANSACTIONS ON MULTIMEDIA