Recognition-Aware Learned Image Compression
This addresses the need for efficient image compression that preserves recognition performance for applications like classification and object detection, representing an incremental improvement over existing learned compression methods.
The paper tackles the problem of learned image compression for downstream recognition tasks by jointly optimizing compression and recognition networks, achieving up to 26% higher recognition accuracy at low bitrates compared to traditional methods like BPG.
Learned image compression methods generally optimize a rate-distortion loss, trading off improvements in visual distortion for added bitrate. Increasingly, however, compressed imagery is used as an input to deep learning networks for various tasks such as classification, object detection, and superresolution. We propose a recognition-aware learned compression method, which optimizes a rate-distortion loss alongside a task-specific loss, jointly learning compression and recognition networks. We augment a hierarchical autoencoder-based compression network with an EfficientNet recognition model and use two hyperparameters to trade off between distortion, bitrate, and recognition performance. We characterize the classification accuracy of our proposed method as a function of bitrate and find that for low bitrates our method achieves as much as 26% higher recognition accuracy at equivalent bitrates compared to traditional methods such as Better Portable Graphics (BPG).