search

msracver / FCIS

Fully Convolutional Instance-aware Semantic Segmentation
MIT License
1.57k stars 415 forks source link

Nan value appears during training with COCO dataset #134

Open w-tingting opened 6 years ago

w-tingting commented 6 years ago

7 When I use the Coco dataset for training, the network encounters an invalid value and and loss falls into nan.Anyone knows how to solve it？Thank you!