About difference between pretrained model's output and my model's output #91
Description
Well, I fixed the trilinear.cpp and some .py files to make work well on recent Pytorch (2.7.0, cu126).
when I run inference python file to get output, I wondered why the outputs look quite different between pretrained one and my model.
I think I didn't fix anything about the things that can affect to output's quality directly.
I used MIT5k dataset to train the model, 4,500 images per epoch for training. and I used 269th checkpoint files to inference the outputs.
also, I set the all parameter values from this paper.
the outputs :
Did you think that I fixed the model wrong? or It's usual that the outputs may be different compared by pretrained model?
Any suggests are welcome.