Table 1, column Without iCNN/DCTIF selection, the average BD-rate of Class F is -0.4% BD-rate compared to HM. Average all sequences which are calculated based on 20 sequences are correct.

BibTeX does not have the right entry for preprints. This is a hack for producing the correct reference:

@Booklet{EasyChair:1398,
author = {Chi Do-Kim Pham and Jinjia Zhou},
title = {iCNN: A Convolutional Neural Network for Fractional Interpolation in Video Coding},
howpublished = {EasyChair Preprint no. 1398},
year = {EasyChair, 2019}}