search

isht7 / pytorch-deeplab-resnet

DeepLab resnet v2 model in pytorch
MIT License
602 stars 117 forks source link

problem of train.py #42

Open kkhuang1990 opened 6 years ago

kkhuang1990 commented 6 years ago

In line 223 of train.py file. input and target should be both out[0] and label[0] if we want to use data after multi-scale fusion for training.

kkhuang1990 commented 6 years ago

Also for esize_label_batch() function in train.py, instead of UpsamplingBilinear2d but UpsamplingNearest2d should be used for resizing labels since labels can only be integer, converting float to int may cause mis-labeling for ground truth labels