Towards Image Understanding from Deep Compression without Decoding
This addresses the problem of computational efficiency for image understanding tasks in applications like storage and bandwidth reduction, but it is incremental as it builds on existing DNN-based compression methods.
The paper tackles the problem of performing image understanding tasks like classification and segmentation directly on compressed representations from DNN-based image compression, bypassing decoding to RGB space. The result shows comparable accuracies to networks on compressed RGB images while reducing computational complexity up to 2×, with synergies from joint training improving image quality and performance, especially at aggressive compression rates.
Motivated by recent work on deep neural network (DNN)-based image compression methods showing potential improvements in image quality, savings in storage, and bandwidth reduction, we propose to perform image understanding tasks such as classification and segmentation directly on the compressed representations produced by these compression methods. Since the encoders and decoders in DNN-based compression methods are neural networks with feature-maps as internal representations of the images, we directly integrate these with architectures for image understanding. This bypasses decoding of the compressed representation into RGB space and reduces computational cost. Our study shows that accuracies comparable to networks that operate on compressed RGB images can be achieved while reducing the computational complexity up to $2\times$. Furthermore, we show that synergies are obtained by jointly training compression networks with classification networks on the compressed representations, improving image quality, classification accuracy, and segmentation performance. We find that inference from compressed representations is particularly advantageous compared to inference from compressed RGB images for aggressive compression rates.