Lossy image compression is often limited by the simplicity of the chosen loss measure. Recent research suggests that generative adversarial networks have the ability to overcome this limitation and serve as a multi-modal loss, especially for textures. Together with learned image compression, these two techniques can be used to great effect when relaxing the commonly employed tight measures of distortion. However, convolutional neural network-based algorithms have a large computational footprint. Ideally, an existing conventional codec should stay in place, ensuring faster adoption and adherence to a balanced computational envelope.As a possible avenue to this goal, we propose and investigate how learned image coding can be used as a surrogate to optimise an image for encoding. A learned filter alters the image to optimise a different performance measure or a particular task. Extending this idea with a generative adversarial network, we show how entire textures are replaced by ones that are less costly to encode but preserve a sense of detail.Our approach can remodel a conventional codec to adjust for the MS-SSIM distortion with over 20% rate improvement without any decoding overhead. On task-aware image compression, we perform favourably against a similar but codec-specific approach.

Deep learning for image and video compression has demonstrated promising results both as a standalone technology and a hybrid combination with existing codecs. However, these systems still come with high computational costs. Deep learning models are typically applied directly in pixel space, making them expensive when resolutions become large.In this work, we propose an online-trained upsampler to augment an existing codec. The upsampler is a small neural network trained on an isolated group of frames. Its parameters are signalled to the decoder. This hybrid solution has a small scope of only 10s or 100s of frames and allows for a low complexity both on the encoding and the decoding side.Our algorithm works in offline and in zero-latency settings. Our evaluation employs the popular x265 codec on several high-resolution datasets ranging from Full HD to 8K. We demonstrate rate savings between 8.6% and 27.5% and provide ablation studies to show the impact of our design decisions. In comparison to similar works, our approach performs favourably.

CNN-based stereo matching methods achieve great performance but come with high computational requirements. Pruning a CNN can reduce the complexity but may in turn lead to memory conflicts, lowering throughput. Our proposed architecture and memory mapping technique aim at reducing conflicts to exploit extremely sparse stereo matching networks. To maintain a high utilization of processing elements, we decompose the de-convolution operation into several convolution operations. The proposed architecture provides a 2.1 x speed up over SCNN. Compared to the software implementation, only 0.01% performance drop is observed, so that the proposed architecture obtains state-of-the-art accuracy compared to existing sparsity aware hardware implementations.

Digital media is ubiquitous and produced in ever-growing quantities. This necessitates a constant evolution of compression techniques, especially for video, in order to maintain efficient storage and transmission. In this work, we aim at exploiting non-local redundancies in video data that remain difficult to erase for conventional video codecs We design convolutional neural networks with a particular emphasis on low memory and computational footprint. The parameters of those networks are trained on the fly, at encoding time, to predict the residual signal from the decoded video signal. After the training process has converged, the parameters are compressed and signalled as part of the code of the underlying video codec. The method can be applied to any existing video codec to increase coding gains while its low computational footprint allows for an application under resource-constrained conditions. Building on top of High Efficiency Video Coding, we achieve coding gains similar to those of pretrained denoising CNNs while only requiring about 1% of their computational complexity Through extensive experiments, we provide insights into the effectiveness of our network design decisions. In addition, we demonstrate that our algorithm delivers stable performance under conditions met in practical video compression: our algorithm performs without significant performance loss on very long random access segments (up to 256 frames) and with moderate performance drops can even be applied to single frames in high-resolution low delay settings.

Scene Parsing is a crucial step to enable autonomous systems to understand and interact with their surroundings. Supervised deep learning methods have made great progress in solving scene parsing problems, however, come at the cost of laborious manual pixel-level annotation. Synthetic data as well as weak supervision have been investigated to alleviate this effort. Nonetheless, synthetically generated data still suffers from severe domain shift while weak labels often lack precision. Moreover, most existing works for weakly supervised scene parsing are limited to salient foreground objects. The aim of this work is hence twofold: Exploit synthetic data where feasible and integrate weak supervision where necessary. More concretely, we address this goal by utilizing depth as transfer domain because its synthetic-to-real discrepancy is much lower than for color. At the same time, we perform weak localization from easily obtainable image level labels and integrate both using a novel contour-based scheme. Our approach is implemented as a teacher-student learning framework to solve the transfer learning problem by generating a pseudo ground truth. Using only depth-based adaptation, this approach already outperforms previous transfer learning approaches on the popular indoor scene parsing SUN RGB-D dataset. Our proposed two-stage integration more than halves the gap towards fully supervised methods when compared to previous state-of-the-art in transfer learning.

CNN-based super-resolution methods achieve great performance but come with high bandwidth requirements. This paper proposes a bandwidth efficient CNN-based architecture for super resolution supporting up to Full HD images at 60 fps. The bandwidth efficiency is achieved by layer fusion with data reuse scheduling and dynamic quantization. To maintain a high utilization of processing elements, the deconvolution operation is decomposed into a several convolution operations. The proposed architecture provides a 94.8 % bandwidth reduction. Compared to the software implementation, less than 1 % performance drop is induced in this VLSI implementation. The enables the proposed architecture to obtain state-of-the-art accuracy compared to existing super-resolution hardware implementations.

Photo composition is an important factor affecting the aesthetics in photography. However, it is a highly challenging task to model the aesthetic properties of good compositions due to the lack of globally applicable rules to the wide variety of photographic styles. Inspired by the thinking process of photo taking, we formulate the photo composition problem as a view finding process which successively examines pairs of views and determines their aesthetic preferences. We further exploit the rich professional photographs on the web to mine unlimited high-quality ranking samples and demonstrate that an aesthetics-aware deep ranking network can be trained without explicitly modeling any photographic rules. The resulting model is simple and effective in terms of its architectural design and data sampling method. It is also generic since it naturally learns any photographic rules implicitly encoded in professional photographs. The experiments show that the proposed view finding network achieves state-of-the-art performance with sliding window search strategy on two image cropping datasets.