Awesome

E-PGN

Tensorflow implementation of 'Episode-based Prototype Generating Network for Zero-Shot Learning'

Preparation

• Prerequisites
  • Tensorflow (r1.4 - r1.12 should work fine)
  • Python 2.7 with matplotlib, numpy and scipy
• Datasets
  • download the model from Google Drive or Baidu Cloud (bphc), and unzip the files to ./data/,

Training & Test

Exemplar commands are listed here for AwA1 dataset.

• You can

  sh ./run_script/1awa.sh
  

• or

   python ../scripts/1awa.py --att_dim 85 --hid_dim 1800 --mid_dim 1200 --cla_num 50 --tr_cla_num 40 --selected_cla_num 10 --lr 5e-5 --epoch 30 --episode 50 --inner_loop 100 --batch_size 100 --dropout --manualSeed 4198