ZSLPP
Mohamed Elhoseiny*, Yizhe Zhu*, Han Zhang, Ahmed Elgammal, Link the head to the "peak'': Zero Shot Learning from Noisy Text descriptions at Part Precision, CVPR, 2017
This code is implemente by Yizhe Zhu and Mohamed Elhoseiny.
Data: You can download the data CUB2011 and NABird.
Trianed Models:
https://drive.google.com/open?id=0B_8vkk7CF-pwMU5QQUlUOTZFblU reproduce the results in the paper.
Raw wikipedia article data and detailed merging information of NABird can be obtained here.
#Testing, reproducing the results in the paper
ZSL_Test(Dataset = 'CUBird' or 'NABird', splitmode = 'East' or 'Hard', ImgFtSource = 'DET' or 'ATN')
splitmode = Easy or Hard splits defined in Section 4.1 in the paper
CUNBirds East split in Table1 (ATN means usin groun truth part bozes)
ZSL_Test('CUBird', 'Easy', 'ATN') Dataset: CUB2011 Easy ATN Model: trained_models/CUBird_Easy_ATN.mat Load Testing set test_acc = 43.5049%
ZSL_Test('CUBird', 'Easy', 'DET'), :DET means using the detected parts instead of GT parts.
Dataset: CUB2011 Easy DET Model: trained_models/CUBird_Easy_DET.mat Load Testing set test_acc = 37.5725%
NABirds Easy split in Table3 :DET means using the detected parts instead of GT parts.
ZSL_Test('NABird', 'Easy') Dataset: NABird Easy DET Model: trained_models/NABird_Easy_DET.mat Load Testing set test_acc = 30.5937%
NABirds Hard split in Table3
ZSL_Test('NABird', 'Hard') Dataset: NABird Hard DET Model: trained_models/NABird_Hard_DET.mat Load Testing set test_acc = 8.1349%
Training
ZSL_Train(Dateset, Splitmode, ImgFtSource, lambda1, lambda2, GPU_mode) is the command to train the model using a particular setting. % For example ZSL_Train('CUBird', 'Easy', 'DET', 100000, 10000, true), trains on the CUBirds dataset on the Easy split and using the detected part boxes. , lambda1=100000, and lambda2=10000, and GPU_mode=true (using GPU mode for training). If false, the training is done on CPU.