Visual and Semantic Alignment Contrastive Network for Generalized Zero-shot Learning

Environment

This code is running at Windows 10 system, with 4*2080TI, the Cuda version is 11.3.

Dependencies

• pip install -r requirements.txt

Datasets

Download APY,AWA1,AWA2,CUB,FLO,SUN from http://datasets.d2.mpi-inf.mpg.de/xian/xlsa17.zip, then unzip the xlsa17.zip and change the directory xlsa17/data to xlsa17/datasets. Finally, move the datasets directory to the project directory ./datasets.

Train and Test

Run python train.py with the corresponding parameter. For example:

python train.py --dataset AWA1 --syn_num 600 --attSize 85 --nz 3 --nclass_all 50 --nclass_seen 40 --gpu 2 --k 147 --begin_step 50 --test_epoch 1 --cr 10 --temp 0.06 --reg 20 --lr 0.0002 --d 0.0001 --nepoch 400 --syn_t 13 

The parameters for other datasets can be found in parameters.txt