Short programming project: Few-shot learning for image classification

The project explore the findings in the paper Revisiting Local Descriptor based Image-to-Class Measure for Few-shot Learning by Wenbin Li, Lei Wang, Jinglin Xu, Jing Huo, Yang Gao and Jiebo Luo. In CVPR 2019.

Prerequisites

• Linux
• Python 3.8 - 3.10
• All library dependencies from requirements.txt

Getting Started

Installation

• Clone this repo:

git clone https://github.com/d33dler/spp.git

• Install requirements.txt dependencies
• cd to the executables folder and run python exec.py --help to see the available options.

Datasets

• miniImageNet.
• StanfordDog.
• StanfordCar.
• CUB-200.
  

For training and testing on different conditions you must edit the model root config. Refer to models/architectures/configs/* for examples

Training

• Train a model based on Conv64F:

python exec.py --arch DN_X --config models/architectures/DN4_Vanilla --dataset_dir your/dataset/path --data_name aName \
--mode train --epochs 30 --dengine --refit_dengine

Testing

• Test the model (omit --dengine to use the original DN4 and omit --refit_dengine to use the pre-trained dengine):

python exec.py --arch DN_X --config models/architectures/DN4_Vanilla --dataset_dir your/dataset/path --data_name aName \
--mode test --dengine --resume your/model/path.pth.tar

• Results:
  

TODOs

• Add config storing in model
• Add more datasets
• Add more architectures
• Add more experiments