Whether the results can be better or worse if we replace the nearest neigbor resize into transpose convolution operation, cause the nearest neigbor is kind of lack of elegance to me.

[yeephycho]

Thank you for this excellent work! If there's any experiments supporting any conclusion of above question, please let us know ：）

[yeephycho]

https://github.com/HRNet/HRNet-Semantic-Segmentation/issues/72