AAVAE
Official implementation of the paper "AAVAE: Augmentation-AugmentedVariational Autoencoders"
Abstract
Recent methods for self-supervised learning can be grouped into two paradigms: contrastive and non-contrastive approaches. Their success can largely be attributed to data augmentation pipelines which generate multiple views of a single input that preserve the underlying semantics. In this work, we introduce augmentation-augmented variational autoencoders (AAVAE), a third approach to self-supervised learning based on autoencoding. We derive AAVAE starting from the conventional variational autoencoder (VAE), by replacing the KL divergence regularization, which is agnostic to the input domain, with data augmentations that explicitly encourage the internal representations to encode domain-specific invariances and equivariances. We empirically evaluate the proposed AAVAE on image classification, similar to how recent contrastive and non-contrastive learning algorithms have been evaluated. Our experiments confirm the effectiveness of data augmentation as a replacement for KL divergence regularization. The AAVAE outperforms the VAE by 30% on CIFAR-10 and 40% on STL-10. The results for AAVAE are largely comparable to the state-of-the-art for self-supervised learning.
Training
To train the AAVAE model
- Create a python virtual environment.
python setup.py install.- Train using
python src/vae.py --denoising.
To reproduce the results from the paper on CIFAR-10:
python src/vae.py \
--gpus 1 \
--max_epochs 3200 \
--batch_size 256 \
--warmup_epochs 10 \
--val_samples 16 \
--weight_decay 0 \
--logscale 0 \
--kl_coeff 0 \
--learning_rate 2.5e-4
To evaluate the pretrained encoder
python src/linear_eval.py --ckpt_path "path\to\saved\file.ckpt"
Saved checkpoints
| Model | Dataset | Checkpoint | Downstream acc. |
|---|---|---|---|
| AAVAE | CIFAR-10 | checkpoint | 87.14 |
| AAVAE | STL-10 | checkpoint | 84.72 |
243 Dec 26, 2022
![[PyTorch] Official implementation of CVPR2021 paper](https://github.com/XuyangBai/PointDSC/raw/master/figures/pipeline.png)
153 Dec 14, 2022
367 Dec 27, 2022
1k Dec 31, 2022
59 Sep 25, 2022
29 Sep 2, 2022
109 Dec 23, 2022
574 Jan 2, 2023
115 Dec 28, 2022
101 Nov 25, 2022
49 Nov 23, 2022
272 Dec 28, 2022
201 Dec 21, 2022
364 Dec 14, 2022
111 Dec 31, 2022
35 Dec 6, 2022
59 Dec 17, 2022
307 Jan 3, 2023