search

fangchangma / sparse-to-dense.pytorch

ICRA 2018 "Sparse-to-Dense: Depth Prediction from Sparse Depth Samples and a Single Image" (PyTorch Implementation)
445 stars 101 forks source link

Question about pretrained model #11

Closed ziqi-zhang closed 6 years ago

ziqi-zhang commented 6 years ago

Hi, thanks for sharing the code. These days I'm trying your code but I found that Lg10=nan(nan) and I didn't make it to train the net. I guess it was I didn't use any pretrained model. I wonder is the pretrained model on ImageNet necessary when using this version?

ziqi-zhang commented 6 years ago

And here is my script python3 main.py \ -a resnet50 \ -d upproj \ -m rgbd \ -s 100 \ --epochs 15 \ --pretrained

fangchangma commented 6 years ago

Hi Ziqi. Lg10=nan is expected since there are negative prediction values, and not using the imagenet-pretrained models is also fine (I just pushed another commit, and please use "--pretrained False" if you don't want to use the pretrained network). What exactly do you mean by "didn't make it to train the net"?

ziqi-zhang commented 6 years ago

Thanks! I was so careless and didn't read the code. Now I have found the mistake and the model is running~