search

torrvision / siamfc-tf

SiamFC tracking in TensorFlow.
568 stars 189 forks source link

when i make tracking, the fps is only 17-18 #19

Open zhenyingfang opened 7 years ago

zhenyingfang commented 7 years ago

use gpu:k40 and tensorflow 1.1.0

jvlmdr commented 7 years ago

Is this with visualization disabled? (https://github.com/torrvision/siamfc-tf/issues/4)

We tested the code on M40 and Titan X. I believe K40 is significantly slower than M40.

Does this speed include time to load the images? It may be faster to keep them on a RAM disk or SSD than on a spinning hard drive.

zhenyingfang commented 7 years ago

in siamese.py line:38: image_file = tf.read_file(filename), Maybe it takes a lot of time?

jvlmdr commented 7 years ago

Yes, maybe.

You can profile TensorFlow code using the method described in this answer: https://stackoverflow.com/a/37774470/1136018

zhenyingfang commented 6 years ago

do you test this tensorflow code, my fps is still 17

mingchuan123 commented 6 years ago

Hi guys, have you get a final result on the fps of the trackers on K40?

fzqneo commented 6 years ago

For your reference, I port the program to Python 3 using 2to3.exe and run it on Windows 10 with GTX 1080, TensorFlow 1.8.0.

FPS w/ visualization = 4 FPS w/o visualization = 25