README.md

Adaptive Dimension Reduction and Variational Inference for Transductive Few-Shot Classification

This repository is the official implementation of [Adaptive Dimension Reduction and Variational Inference for Transductive Few-Shot Classification].

Requirements

To install requirements:

pip install -r requirements.txt

Evaluation

To evaluate my model on CUB for the unbalanced setting, run:

python bavardage.py --shot 1/5 --T_km 10 --T_vb 4 --clp 5

For the balanced setting, run:

python bavardage.py --shot 1/5 --T_km 10 --T_vb 4 --clp 5 --balanced

Pre-trained Models

We provide a WRN model pretrained on CUB in ./checkpoints

Results

Our model achieves the following performance on :

CUB

Model name	1 shot (unbalanced)	5 shot (unbalanced)	1 shot (balanced)	5 shot (balanced)
BAVARDAGE	82.00%	90.67%	85.64%	91.42%