https://github.com/nimavahdat/few-shot-learning

This is the implementation of the approach described in the paper "Few-Shot Learning via Embedding Adaptation with Set-to-Set Functions" by Ye, Han-Jia et al. The proposed approach adapts instance embeddings to the target classification task with a set-to-set function and achieves state-of-the-art results on multiple few-shot learning benchmarks.
https://github.com/nimavahdat/few-shot-learning

Last synced: about 11 hours ago
JSON representation

This is the implementation of the approach described in the paper "Few-Shot Learning via Embedding Adaptation with Set-to-Set Functions" by Ye, Han-Jia et al. The proposed approach adapts instance embeddings to the target classification task with a set-to-set function and achieves state-of-the-art results on multiple few-shot learning benchmarks.

• Host: GitHub
• URL: https://github.com/nimavahdat/few-shot-learning
• Owner: NimaVahdat
• License: mit
• Created: 2023-04-07T17:51:00.000Z (over 1 year ago)
• Default Branch: main
• Last Pushed: 2024-09-14T01:25:58.000Z (about 2 months ago)
• Last Synced: 2024-09-14T14:59:34.875Z (about 2 months ago)
• Language: Python
• Homepage:
• Size: 467 KB
• Stars: 1
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# FEAT: Few-Shot Learning via Embedding Adaptation with Set-to-Set Functions

This is the implementation of the approach described in the paper "Few-Shot Learning via Embedding Adaptation with Set-to-Set Functions" by Ye, Han-Jia et al. The proposed approach adapts instance embeddings to the target classification task with a set-to-set function and achieves state-of-the-art results on multiple few-shot learning benchmarks.

## Installation

Required packages:

* Pytorch

* tensorboardX

* Numpy

 

## Dataset

Once you have downloaded the dataset, you will need to create a new folder named "images" within the "minimagenet" or "retail" folder, and place all of the images into this folder. The data loader that we have provided will automatically read the images from the "images" folder.

## Training

Example for traing the model

```bash

python train.py --lr 0.0001 --temperature 64   \

--max_epoch 100 --model_type AmdimNet --dataset MiniImageNet \

--init_weights ./miniimagenet.pth  \

--save_path ./MINI_1shot_5way/ \

--shot 1  --way 5 --step_size 10 --gamma 0.5 \

```

The results on the MiniImageNet and TieredImageNet datasets are shown below:

## MiniImageNet

| Model | 1-Shot 5-Way	| 5-Shot 5-Way |

|:------:|:-------------:|:------------:|

| ProtoNet |	62.21 |	80.64 |

| BILSTM |	63.04 |	80.63 |

| DEEPSETS |	64.24 |	80.51 |

| GCN |	63.93 |	81.65 |

| FEAT |	**66.08** |	**81.95** |

## TieredImageNet

| Model | 1-Shot 5-Way	| 5-Shot 5-Way |

|:------:|:-------------:|:------------:|

| ProtoNet |	67.93 |	84.23 |

| BILSTM |	67.84 |	83.53 |

| DEEPSETS |	68.89 |	84.86 |

| GCN	| 66.20 |	84.64 |

| FEAT |	**70.23** |	**84.37** |

### References

   Ye, Han-Jia, et al. "Few-shot learning via embedding adaptation with set-to-set functions." Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2020.