Lossy Compression for Lossless Prediction
Most data is automatically collected and only ever "seen" by algorithms. Yet, data compressors preserve perceptual fidelity rather than just the information needed by algorithms performing downstream tasks. In this paper, we characterize the bit-rate required to ensure high performance on all predictive tasks that are invariant under a set of transformations, such as data augmentations. Based on our theory, we design unsupervised objectives for training neural compressors. Using these objectives, we train a generic image compressor that achieves substantial rate savings (more than $1000\times$ on ImageNet) compared to JPEG on 8 datasets, without decreasing downstream classification performance.
PDF Abstract NeurIPS 2021 PDF NeurIPS 2021 AbstractCode
Colab
CIFAR-10
ImageNet
MS COCO
MNIST
STL-10
Stanford Cars
Food-101
Caltech-101
PCam
Oxford-IIIT Pet Dataset
Results from the Paper
Ranked #1 on
Image Compression
on Oxford-IIIT Pet Dataset
(using extra training data)
| Task | Dataset | Model | Metric Name | Metric Value | Global Rank | Uses Extra Training Data |
Benchmark |
|---|---|---|---|---|---|---|---|
| Image Compression | Caltech101 | Lossyless Compressor | Bit rate | 1340 | # 1 | ||
| Image Compression | Cars-196 | Lossyless Compressor | Bit rate | 1470 | # 1 | ||
| Image Compression | CIFAR-10 | Lossyless Compressor | Bit rate | 1410 | # 1 | ||
| Image Compression | Food-101 | Lossyless Compressor | Bit rate | 1270 | # 1 | ||
| Image Compression | Oxford-IIIT Pet Dataset | Lossyless Compressor | Bit rate | 1210 | # 1 | ||
| Image Compression | PCam | Lossyless Compressor | Bit rate | 1490 | # 1 | ||
| Image Compression | STL-10 | Lossyless Compressor | Bit rate | 1340 | # 1 |
