search

StevenLiuWen / ano_pred_cvpr2018

Official implementation of Paper Future Frame Prediction for Anomaly Detection -- A New Baseline, CVPR 2018
433 stars 126 forks source link

flownet input / aspect ratio #24

Closed wr0112358 closed 5 years ago

wr0112358 commented 5 years ago

The input to the two flownet models is distorted from aspect ratio 1:1(256x256) to 384x512. I understand that flownet needs inputs sizes dividable by 64. I am trying to understand the reason for this? Why not just pad with zeros to avoid the distortion?

https://github.com/StevenLiuWen/ano_pred_cvpr2018/blob/master/Codes/models.py#L24

StevenLiuWen commented 5 years ago

@wr0112358 Here is the implementation of FlowNet that we use, and in its implementation, the input size must be 384 x 512 (or (k 192) x (k 256)). For convenience, we resize the image from 256 x 256 to 384 x 512.