dummy-link

jlimgnet

process imgnet features using julia

Readme

reports

Dataset

we randomly choose 100 imagenet classes under the condition that each class has 800~1200 images.

use LIBLINEAR directly

for a 50/50 split of train/test, we have svm accuracy: 76.29%. for the first 10 classes, the same 50/50 split, accuracy is: 92.55%.

shape of data

average radius of classes:

average distances between center of classes

First Commit

09/10/2015

Last Touched

over 5 years ago

Commits

3 commits