I wonder if this model could handle any size of input or resize into 224x224 as input? about fast-depth

[dwofk]

Hi @Hackerlil

This model requires a 224x224 input to the first layer. Resizing a larger input to 224x224 would work.

[Hackerlil]

ok, another question,the nyudepthv2 val folder,the raw hdf5 file contains depth files,the depth info is between about 0.0 and 10.0, I want to save it for other format,such as 0-255 gray or .pgm files,means normalize.can it works?

[fangchangma]

@Hackerlil Not sure if we follow your question perfectly. Turning a float-point representation into 8-bit char should work in principle (with proper scaling), but at the cost of precision loss. It also requires retraining of the networks.