About another dataset

[rrxi]

I want to use CrevNet for abnormal event detection of video. I convert Avenue dataset to 256x256x1 according to convert_kth.lua, get this error

then, I convert avenue to 64x64x1, it can work, but get this result

I want to know, which training parameters need to be modified.

[rrxi]

My solution is to change param w, h of zig_rev_predictor.

• image_size=64 => w=h=8
• image_size=128 => w=h=16
• image_size=256 => w=h=32

I am not sure if this modification will affect performance.