https://github.com/NVIDIA/DIGITS
Revision b4f326186f5f02b6fd08e19a771735298808e7b3 authored by Luke Yeager on 22 April 2015, 20:33:40 UTC, committed by Luke Yeager on 22 April 2015, 20:33:40 UTC
"half_crop" is not used in any deep learning research and should not be
chosen as the default. Also, for some datasets the network can simply
learn the shape of the noise (e.g. one category may be bottles, so the
network learns that all tall and thin images with noise on the left and
right is a bottle)
1 parent 628496d
History
Tip revision: b4f326186f5f02b6fd08e19a771735298808e7b3 authored by Luke Yeager on 22 April 2015, 20:33:40 UTC
Change default resize mode to squash
Tip revision: b4f3261
File Mode Size
digits
docs
tools
.gitignore -rw-r--r-- 55 bytes
LICENSE -rw-r--r-- 1.5 KB
README.md -rw-r--r-- 2.8 KB
digits-devserver -rwxr-xr-x 1.5 KB
digits-server -rwxr-xr-x 282 bytes
digits-test -rwxr-xr-x 283 bytes
digits-walkthrough -rwxr-xr-x 5.9 KB
gunicorn_config.py -rw-r--r-- 479 bytes
nginx.site -rw-r--r-- 1.0 KB
requirements.txt -rw-r--r-- 220 bytes

README.md

back to top