-
Notifications
You must be signed in to change notification settings - Fork 0
/
README
28 lines (22 loc) · 1.25 KB
/
README
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
Dependencies:
-------------
The code uses MOSEK to solve QP and LP optimizations. Please download
the MOSEK software for Linux (http://www.mosek.com/resources/downloads)
along with a mosek license (academic license is free: http://license.mosek.com/academic/).
Then set the corresponding directory and license paths in optimizerSetup.m.
Data:
-----
Download the data (http://vision.stanford.edu/vigneshr_release_data/eccv14_episode_data.zip)
and unpack it in a directory $DATA_DIR.
Running the code:
-----------------
1. The bidriectional optimization can be run by running "runCorefFaceOpt".
The different arguments are explained in the file.
2. Results can be generated for the provided dataset by running main($DATA_DIR).
Saved results:
--------------
1. While running "runCorefFaceOpt.m", results are saved to the directory "$DATA_DIR/$EPISODE_NAME/bidirectional_reulsts/"
2. The face recognition results are saved at each iteration in the files faces_Y.*, and similarly
coref results are stored for each iteration in the file faces_Z.*.
3. The coref results are stored in a matrix Z_whole as a matrix [0,1]^{number_of_cast_names X number_of_mentions}.
4. The face results are stored in a matrix Y_whole as a matirx [0,1]^{number_of_cast_names X number_of_faces}.