Skip to content

Unofficial implement of LEAP(Deep Representation Learning on Long-tailed Data: A Learnable Embedding Augmentation Perspective) for Multi-Label Classification.

Notifications You must be signed in to change notification settings

stxupengyu/CVPR-2020-LEAP

Repository files navigation

Deep Representation Learning on Long-tailed Data: A Learnable Embedding Augmentation Perspective

Unofficial implement for Multi-Label Classification.

Reproducibility:

python main.py --dataset AAPD python main.py --dataset RCV python main.py --dataset Eurlex

configuration file

Please confirm the corresponding configuration file: config_AAPD config_RCV config_Eurlex

configuration file describe

epochs: 10
lstm_hidden_dimension: 150
batch_size: 64
d_a: 100 #attention dim
emb_size: 300
GPU: True
GPU_Number: 2
load_path: '/data/dataset/RCV/'#data path
data_token: ''#no use in current code
lr: 0.001
scale: 120 # para of CosFace Loss
quantile: 0.8 #tail label ratio
gamma: 0.1 #center learning rate for LEAP

About

Unofficial implement of LEAP(Deep Representation Learning on Long-tailed Data: A Learnable Embedding Augmentation Perspective) for Multi-Label Classification.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages