Source code of Transfer Increment for Generalized Zero-Shot Learning In TNNLS 2021
The details of model can be found in
L. J. Feng, et al. Transfer Increment for Generalized Zero-Shot Learning, TNNLS, 2021.
Link: https://pan.baidu.com/s/12SGtog43jJ0cZRpGOqTgxg
code: vevv
pip3 install cvxpy
python3 AWA2_PA.py
============================Experiment: Ours+IOM+Res.============================
(24270, 2048) (24270,) (24270, 85)
(6985, 2048) (6985,) (6985, 85)
(6067, 2048) (6067,) (6067, 85)
Base training for 85 attribute classifiers begin...
10 th classifier is training
20 th classifier is training
30 th classifier is training
40 th classifier is training
50 th classifier is training
60 th classifier is training
70 th classifier is training
80 th classifier is training
85 th classifier is training
Base training for 85 attribute classifiers is over！
Incremental training begin...
Incremental samples are 1, S is 0.8757, U is 0.1907, H is 0.3132, training time is 10.897.
Incremental samples are 3, S is 0.8698, U is 0.2261, H is 0.3588, training time is 10.6462.
Incremental samples are 5, S is 0.8584, U is 0.3173, H is 0.4633, training time is 10.3452.
Incremental samples are 7, S is 0.8121, U is 0.3812, H is 0.5189, training time is 10.7549.
Incremental samples are 9, S is 0.8142, U is 0.4212, H is 0.5552, training time is 10.0988.
Incremental samples are 10, S is 0.8388, U is 0.4467, H is 0.5829, training time is 11.7884.
Incremental samples are 13, S is 0.8108, U is 0.4422, H is 0.5723, training time is 10.9359.
Incremental samples are 15, S is 0.7872, U is 0.48, H is 0.5964, training time is 9.9649.
Incremental samples are 17, S is 0.7524, U is 0.5227, H is 0.6169, training time is 10.8031.
Incremental samples are 19, S is 0.7898, U is 0.5162, H is 0.6244, training time is 10.2848.
Incremental samples are 20, S is 0.7391, U is 0.5609, H is 0.6378, training time is 10.1297.
Incremental samples are 21, S is 0.7551, U is 0.5288, H is 0.622, training time is 9.6607.
Incremental samples are 23, S is 0.7646, U is 0.5658, H is 0.6503, training time is 9.9159.
Incremental samples are 25, S is 0.7493, U is 0.5684, H is 0.6464, training time is 9.9403.
Incremental samples are 27, S is 0.7242, U is 0.6089, H is 0.6616, training time is 10.0022
Incremental samples are 29, S is 0.6816, U is 0.6429, H is 0.6617, training time is 10.3731.
Incremental samples are 30, S is 0.6855, U is 0.5973, H is 0.6384, training time is 10.035.
Incremental samples are 31, S is 0.7073, U is 0.591, H is 0.6439, training time is 10.2678.
Incremental samples are 33, S is 0.685, U is 0.6462, H is 0.6651, training time is 10.4913.
Incremental samples are 35, S is 0.6903, U is 0.5827, H is 0.6319, training time is 10.2225.
Incremental samples are 37, S is 0.686, U is 0.6445, H is 0.6646, training time is 10.3919.
Incremental samples are 39, S is 0.6718, U is 0.6252, H is 0.6477, training time is 10.5959.
Incremental samples are 40, S is 0.673, U is 0.6014, H is 0.6352, training time is 10.1056.
Incremental samples are 43, S is 0.6397, U is 0.6736, H is 0.6562, training time is 10.6409.
Incremental samples are 45, S is 0.6338, U is 0.6588, H is 0.6461, training time is 10.0113.
Incremental samples are 47, S is 0.7005, U is 0.6377, H is 0.6676, training time is 10.2222.
Incremental samples are 50, S is 0.66, U is 0.6704, H is 0.6652, training time is 11.9292.
Incremental samples are 55, S is 0.6438, U is 0.6474, H is 0.6456, training time is 12.7336.
Incremental samples are 60, S is 0.6572, U is 0.674, H is 0.6655, training time is 13.9888.
[1] L. J. Feng, et al. Fault Description Based Attribute Transfer for Zero-Sample Industrial Fault Diagnosis, TII, 2021.
[2] L. J. Feng, et al. Transfer Increment for Generalized Zero-Shot Learning, TNNLS, 2021.