Long-tail learning with class descriptors
6 papers with code • 4 benchmarks • 4 datasets
Long-tail learning by using class descriptors (like attributes, class embedding, etc) to learn tail classes as well as head classes.
Benchmarks
These leaderboards are used to track progress in Long-tail learning with class descriptors
| Trend | Dataset | Best Model | Paper | Code | Compare |
|---|---|---|---|---|---|
|
|||||
|
|||||
|
|||||
|
CUB-200-2011
AwA
ImageNet-LT
Most implemented papers
Learning Imbalanced Datasets with Label-Distribution-Aware Margin Loss
kaidic/LDAM-DRW
•
•
NeurIPS 2019
Deep learning algorithms can fare poorly when the training dataset suffers from heavy class-imbalance but the testing criterion requires good generalization on less frequent classes.
Decoupling Representation and Classifier for Long-Tailed Recognition
facebookresearch/classifier-balancing
•
•
ICLR 2020
The long-tail distribution of the visual world poses great challenges for deep learning based classification models on how to handle the class imbalance problem.
Large-Scale Long-Tailed Recognition in an Open World
zhmiao/OpenLongTailRecognition-OLTR
•
•
CVPR 2019
We define Open Long-Tailed Recognition (OLTR) as learning from such naturally distributed data and optimizing the classification accuracy over a balanced test set which include head, tail, and open classes.
Generalized Zero- and Few-Shot Learning via Aligned Variational Autoencoders
edgarschnfld/CADA-VAE-PyTorch
•
•
CVPR 2019
Many approaches in generalized zero-shot learning rely on cross-modal mapping between the image feature space and the class embedding space.
From Generalized zero-shot learning to long-tail with class descriptors
Real-world data is predominantly unbalanced and long-tailed, but deep models struggle to recognize rare classes in the presence of frequent classes.
Parameter-Efficient Long-Tailed Recognition
shijxcs/pel
•
•
In this paper, we propose PEL, a fine-tuning method that can effectively adapt pre-trained models to long-tailed recognition tasks in fewer than 20 epochs without the need for extra data.
