Meta-Baseline: Exploring Simple Meta-Learning for Few-Shot Learning (ICCV'2021)

Abstract

Meta-learning has been the most common framework for few-shot learning in recent years. It learns the model from collections of few-shot classification tasks, which is believed to have a key advantage of making the training objective consistent with the testing objective. However, some recent works report that by training for whole-classification, i.e. classification on the whole label-set, it can get comparable or even better embedding than many meta-learning algorithms. The edge between these two lines of works has yet been underexplored, and the effectiveness of meta-learning in few-shot learning remains unclear. In this paper, we explore a simple process: meta-learning over a whole classification pre-trained model on its evaluation metric. We observe this simple method achieves competitive performance to state-of-the-art methods on standard benchmarks. Our further analysis shed some light on understanding the trade-offs between the meta-learning objective and the whole-classification objective in few-shot learning. Our code is available at https://github.com/yinboc/few-shot-meta-baseline.

Citation

@inproceedings{chen2021meta,
    title={Meta-Baseline: Exploring Simple Meta-Learning for Few-Shot Learning},
    author={Chen, Yinbo and Liu, Zhuang and Xu, Huijuan and Darrell, Trevor and Wang, Xiaolong},
    booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
    pages={9062--9071},
    year={2021}
}

How to Reproduce Meta Baseline

It consists of three steps:

• Step1: Baseline Base training
  • use all the images of base classes to train a base model with linear head.
  • conduct meta testing on validation set to select the best model.
• Step2: Meta Baseline Base training
  • use all the images of base classes to train a base model with meta metric.
  • conduct meta testing on validation set to select the best model.
• Step3: Meta Testing:
  • use best model from step1.

An example of CUB dataset with Conv4

# baseline base training
python ./tools/classification/train.py \
  configs/classification/baseline/cub/baseline_conv4_1xb64_cub_5way-1shot.py

# Meta Baseline base training
python ./tools/classification/train.py \
  configs/classification/meta_baseline/cub/meta-baseline_conv4_1xb100_cub_5way-1shot.py \
  --options load_from="work_dir/baseline_conv4_1xb64_cub_5way-1shot/best_accuracy_mean.pth"

# meta testing
python ./tools/classification/test.py \
  configs/classification/meta_baseline/cub/meta-baseline_conv4_1xb100_cub_5way-1shot.py \
  work_dir/meta-baseline_conv4_1xb100_cub_5way-1shot/best_accuracy_mean.pth

Note:

• All the result are trained with single gpu.
• The configs of 1 shot and 5 shot use same training setting, but different meta test setting on validation set and test set.
• Currently, we use model selected by 1 shot validation (100 episodes) to evaluate both 1 shot and 5 shot setting on test set.
• The hyper-parameters in configs are roughly set and probably not the optimal one so feel free to tone and try different configurations. For example, try different learning rate or validation episodes for each setting. Anyway, we will continue to improve it.
• The training batch size is calculated by num_support_way * (num_support_shots + num_query_shots)

Results on CUB dataset with 2000 episodes

Results on Mini-ImageNet dataset with 2000 episodes

Results on Tiered-ImageNet dataset with 2000 episodes