OrgLion-ML
/
openface
mirror of https://github.com/cmusatyalab/openface.git
1
0
Fork 0
Code Issues Releases Wiki Activity

Docs: Update links to other projects in 'accuracy'

This commit is contained in:
Brandon Amos 2015-12-02 03:28:53 -05:00
parent cf49f815c6
commit 6742ef326f
1 changed files with 18 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

33
docs/accuracy.md
View File

@ -33,19 +33,22 @@ in `./data/lfw/raw` and `./data/lfw/deepfunneled`.
If you're interested in higher accuracy open source code, see:
1. [Oxford's VGG Face Descriptor](http://www.robots.ox.ac.uk/~vgg/software/vgg_face/),
which is licensed for non-commercial research purposes.
They've released their softmax network, which obtains .9727 accuracy
on the LFW and will release their triplet network (0.9913 accuracy)
and data soon.
## [Oxford's VGG Face Descriptor](http://www.robots.ox.ac.uk/~vgg/software/vgg_face/)
Their softmax model doesn't embed features like FaceNet,
which makes tasks like classification and clustering more difficult.
Their triplet model hasn't yet been released, but will provide
embeddings similar to FaceNet.
The triplet model will be supported by OpenFace once it's released.
2. [AlfredXiangWu/face_verification_experiment](https://github.com/AlfredXiangWu/face_verification_experiment),
which uses Caffe and doesn't yet have a license.
The accuracy on the LFW is .9777.
This model doesn't embed features like FaceNet,
which makes tasks like classification and clustering more difficult.
This is licensed for non-commercial research purposes.
They've released their softmax network, which obtains .9727 accuracy
on the LFW and will release their triplet network (0.9913 accuracy)
and data soon (?).
Their softmax model doesn't embed features like FaceNet,
which makes tasks like classification and clustering more difficult.
Their triplet model hasn't yet been released, but will provide
embeddings similar to FaceNet.
The triplet model will be supported by OpenFace once it's released.
## [AlfredXiangWu/face_verification_experiment](https://github.com/AlfredXiangWu/face_verification_experiment)
This uses Caffe and doesn't yet have a license.
The accuracy on the LFW is .9777.
This model doesn't embed features like FaceNet,
which makes tasks like classification and clustering more difficult.