search

alexgkendall / caffe-segnet

Implementation of SegNet: A Deep Convolutional Encoder-Decoder Architecture for Semantic Pixel-Wise Labelling
http://mi.eng.cam.ac.uk/projects/segnet/
Other
1.08k stars 452 forks source link

Slow inference time #61

Open jboys opened 8 years ago

jboys commented 8 years ago

I followed the SegNet Tutorial and trained SegNet (not SegNet-Basic) on KITTI images (centre-cropped to 360x480) rather than CamVid.

The accuracy was very good, however I am unsure about the inference time. The per-frame forward pass takes on average 227 ms, whereas it was reported to take only 94 ms (for 500x500) in the paper (see table 5): http://arxiv.org/abs/1511.00561

My GPU is a Titan X.

Was there any "trick" used to attain the 94 ms inference time?

If not, does anyone have any ideas on what could be causing the slowdown?

alexgkendall commented 8 years ago

Yes, this timing was using cudnn 5 in Torch. Sorry for the confusion!

Alex

codecolony commented 8 years ago

Hi Alex,

Executing the script to obtain segmentation of input image is taking 14 seconds for me on Titan X. What am I doing wrong? Is there a way to speed this up? Thanks.