MAE-FS

Masked Autoencoders for Few-Shot Learning (MAE-FS) is a self-supervised, generative technique that reinforces few-shot classification performance for a frozen prototypical backbone model.

Experiments

Test MAE-FS on a wide range of out-of-domain and in-domain datasets.

Requirements

Install the requirements necessary for running MAE-FS. A virtual environment (or similar) is recommended.

pip install -r requirements.txt

Datasets

Datasets used in this work can be obtained at the following links:

• mini-imagenet
• BCCD WBC
• HEp-2
• IKEA-FS
• CVPRL2ID Track 1 Small Scale

After downloading all datasets, you should extract/place each respective dataset's folder in the same directory.

Pretrained Weights

You can download pretrained weights for MAE-FS from HuggingFace

To use the pretrained weights, create a weights folder at the root of this repo and place all pretrained weights in it.

Running Experiments

Run the test.sh shell script and specify the number of shots to test and the path to the root data directory (where you placed the downloaded datasets earlier). A 5-shot example run is shown below with a data root of ~/Data:

./test.sh 5 "~/Data"

Custom Experiments

To run an experiment with custom hyperparameters and/or with specific datasets, the fs_test.py can be run directly with arguments applied.

To see a full list of available options, run the following command to see the help dialogue:

python3 fs_test.py --help

Results

In-Domain and Out-of-Domain Results

	In-domain		Out-of-domain
	MIN		BCCD		HEp-2		IKEA-FS
	5w1s	5w5s	5w1s	5w5s	5w1s	5w5s	5w1s	5w5s
MAE-FS (IN1K, ViT-Small)	95.73	98.91	54.24	66.52	58.6	71.21	59.04	76.02
MAE-FS (IN1K, ResNet-18)	88.36	97.11	45.21	59.81	45.02	58.78	51.12	72.89
MAE-FS (MIN, CONV4)	53.86	69.29	46.58	57.53	44.9	56.06	39.23	57.46

CVPR L2ID

	ChestX		ISIC		EuroSAT		Plant Disease
	5w1s	5w5s	5w1s	5w5s	5w1s	5w5s	5w1s	5w5s
MAE-FS (IN1K, ViT-Small)	23.85	27.98	37.09	51.02	75.91	88.77	83.74	95.04
MAE-FS (IN1K, ResNet-18)	21.93	24.12	33.17	46.4	65.54	80.56	78.92	92.58
MAE-FS (MIN, CONV4)	21.73	24.05	33.88	43.44	59.72	72.76	65.57	83.58