Recognition-Aware Learned Image Compression

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 various tasks such classification, object detection, and superresolution. We propose recognition-aware learned method, which optimizes loss alongside task-specific jointly recognition networks. augment hierarchical autoencoder-based network with EfficientNet model use two hyperparameters trade between distortion, bitrate, performance. characterize the classification accuracy of our proposed method function bitrate find that low bitrates achieves much 26% higher at equivalent compared traditional Better Portable Graphics (BPG).