Closed murdockhou closed 5 years ago
Hi,
the HRNet model halves and de-halves the input/the feature maps 5 consecutive times inside the network so the input width and height must be divisible by 2^5 = 32
.
Therefore, you can not use an input size of 192 x 144
, but you can use 192 x 160
or 192 x 128
.
That said, official weights are obtained training the model with input size 256x192 or 384x288 only. With those resolutions the accuracy should be higher.
Thanks for your reply, got it!
Hi, i'm tring to using hrnet with different input size, but there is wrong when I set input size is 192x144, seems like this model can only handle with input 256x192 and 384x288. Is there anyway should to make this work done? Thanks~