Different implementations are described, particularly implementations for video encoding and decoding are presented. According to an implementation, in a method for encoding or decoding a part of an image, an inter-prediction refinement of the image block using optical flow based on boundary smoothed motion compensation is performed. The inter-prediction refinement of the image block further comprises obtaining a motion information for the block, a motion information for the top neighboring block, a motion information for the left neighboring block; and applying an optical flow based on a weighted sum of the obtained motion information to refine the prediction for the block. Advantageously, the refined inter-prediction is applied on boundary sub-blocks of the causal border of the image block or on any sub-block of the image block at a sub-block level if the image block has sub-block motion field.