search

qibao77 / LFFN

Tensorflow code for our paper "Lightweight Feature Fusion Network for Single Image Super-Resolution" (SPL2019)
16 stars 4 forks source link

The question about running time #4

Open minghongli233 opened 4 years ago

minghongli233 commented 4 years ago

Hi, this is a nice work. I want to know if it will increase the training time and inference time obviously by using the multi-path way to extract the features. In addition, I want to know how much more time it will cost than one-path way. Because I use the other multi-path way, it will cost too many time whether in training phase or testing phase. Thank you!

qibao77 commented 4 years ago

Thank you for your attention! We do not compare the inference time between the multi-path method and the single-path method, but only compare the time of the whole LFFN with other mehtods in our paper. However, in my opinion, the multi-path way has the disadvantage of unfriendly memory access according to ShuffleNet-V2 [1], so it may take more time. [1] Ma N, Zhang X, Zheng H T, et al. Shufflenet v2: Practical guidelines for efficient cnn architecture design[J]. arXiv preprint arXiv:1807.11164, 2018.

minghongli233 commented 4 years ago

Thank you for your answer!