Search results

This webinar will present the presenters work on end-to-end learned video compression, with a focus on temporal context mining. Their approach uniquely propagates both reconstructed frames and intermediate features to extract multi-scale temporal contexts, which are then effectively reused across the compression pipeline—including the contextual encoder-decoder, frame generator, and temporal context encoder. By replacing the computationally heavy auto-regressive entropy model, our codec achieves practical encoding/decoding speeds while delivering superior compression performance.

The Multi-Scale Spectral (MSS) loss is widely used for comparing audio signals, offering a good balance between temporal and spectral resolution, while allowing for phase differences between waveforms that are perceptually irrelevant. However, the configuration of this loss function, including parameters such as window type and size, hop size, and magnitude compression, is often chosen empirically and without explicitly considering the impact on loss behavior.

Deep learning models achieve state-of-the-art performance in image restoration but remain largely black-box, making it difficult to understand the role of their weights or adapt them to new settings. Algorithm unrolling offers a more principled mechanism to neural network architecture design. In this talk, the presenters will introduce a specific approach to algorithm unrolling via direct parameterization, the Convolutional Dictionary Learning Network (CDLNet).

Radio frequency fingerprint identification (RFFI) is a promising device authentication technique based on transmitter hardware impairments. The device-specific hardware features can be extracted at the receiver by analyzing the received signal and used for authentication. In this talk, we will present a scalable and channel-robust RFFI framework achieved by deep learning powered radio frequency fingerprint (RFF) extractor and channel independent features.

Modern data analysis and processing tasks typically involve large sets of graph structured data, where the structure carries critical information. Typically, graphs are used as mathematical tools to describe the interactions between the different entities in the system. Characterizing the meso-scale organization, i.e. the community structure, is an important task in network science. Community detection aims to partition the network into sets of nodes that are densely connected internally but sparsely connected to other dense sets of nodes.

Given its remarkable capability for feature extraction in computer vision tasks, deep learning (DL) has been extensively utilized to fuse infrared and visible images. However, the existing DL-based methods generally extract complementary information from source images through convolutional operations, which results in limited preservation of global features. To this end, we propose a novel infrared and visible image fusion method, i.e., the Y-shape dynamic Transformer (YDTR). Speciï¬

Future wireless communication systems are expected to explore spectral bands typically used by radar systems, in order to overcome spectrum congestion of traditional communication bands. Since, in many applications, radar and communication share the same platform, spectrum sharing can be facilitated by joint design as a dual-function radar-communications system.

Multimodal medical image fusion, an effective way to merge the complementary information in different modalities, has become a significant technique to facilitate clinical diagnosis and surgical navigation. However, existing deep fusion models generally depend on convolutional operations, which fails to preserve global context information. To compensate for this defect and achieve accurate fusion, we propose a multiscale adaptive Transformer to fuse multimodal medical images termed MATR.

Multiobject tracking (MOT) is a key challenge in a wide range of applications, such as autonomous navigation and applied ocean sciences. MOT is complicated by object appearance and disappearance, data association ambiguities, and occlusion. Conventional Bayesian methods for MOT, e.g., methods based on belief propagation (BP), entirely rely on a statistical model. This fully model-based approach can lead to highly suboptimal estimates when there is a mismatch between the statistical model and the true data-generating process.

Diffusion models excel at generating photorealistic images, yet their outputs betray a hidden signature: spectral artefacts — distinct anormalies in the frequency domain that deviate from natural image statistics. This webinar will unpack the mathematical underpinnings behind these artefacts, revealing why and how incorrect frequencies emerge during the generation process. We will discuss the extent to which frequency artefacts can be exploited as reliable markers to identify deepfakes, offering new tools for image forensics.