msracver / FCIS

Fully Convolutional Instance-aware Semantic Segmentation
MIT License
1.57k stars 414 forks source link

Speed of training #127

Open michaelhuang74 opened 6 years ago

michaelhuang74 commented 6 years ago

I have two Nvidia K40 GPUs. No matter using coco dataset or VOC dataset, the print out is like follows.

Epoch[0] Batch [100] Speed: 3.00 samples/sec

Is is typical of the training speed of using two K40 GPUs? What exactly does samples/sec mean? One sample is one image?

BTW, I haven't successfully trained on either dataset. For the coco dataset, I encountered the same error as h5py unable to open file (https://github.com/msracver/FCIS/issues/88).