We introduce ShapeWords1, an approach for synthesizing images based on 3D shape guidance and text prompts. ShapeWords incorporates target 3D shape information within specialized tokens embedded together with the input text, effectively blending 3D shape awareness with textual context to guide the image synthesis process. Unlike conventional shape guidance methods that rely on depth maps restricted to fixed viewpoints and often overlook full 3D structure or textual context, ShapeWords generates diverse yet consistent images that reflect both the target shape’s geometry and the textual description. Experimental results show that ShapeWords produces images that are more text-compliant, aesthetically plausible, while also maintaining 3D shape awareness.

We present GEOPARD, a transformer-based architecture for predicting articulation from a single static snapshot of a 3D shape. The key idea of our method is a pretraining strategy that allows our transformer to learn plausible candidate articulations for 3D shapes based on a geometric-driven search without manual articulation annotation. The search automatically discovers physically valid part motions that do not cause detachments or collisions with other shape parts. Our experiments indicate that this geometric pretraining strategy, along with carefully designed choices in our transformer architecture, yields state-of-the-art results in articulation inference in the PartNet-Mobility dataset.

We present ANISE, a method that reconstructs a 3D shape from partial observations (images or sparse point clouds) using a part-aware neural implicit shape representation. The shape is formulated as an assembly of neural implicit functions, each representing a different part instance. In contrast to previous approaches, the prediction of this representation proceeds in a coarse-to-fine manner. Our model first reconstructs a structural arrangement of the shape in the form of geometric transformations of its part instances. Conditioned on them, the model predicts part latent codes encoding their surface geometry. Reconstructions can be obtained in two ways: (i) by directly decoding the part latent codes to part implicit functions, then combining them into the final shape; or (ii) by using part latents to retrieve similar part instances in a part database and assembling them in a single shape. We demonstrate that, when performing reconstruction by decoding part representations into implicit functions, our method achieves state-of-the-art part-aware reconstruction results from both images and sparse point clouds. When reconstructing shapes by assembling parts retrieved from a dataset, our approach significantly outperforms traditional shape retrieval methods even when significantly restricting the database size. We present our results in well-known sparse point cloud reconstruction and single-view reconstruction benchmarks.

Abstract We present a deep learning method that propagates point‐wise feature representations across shapes within a collection for the purpose of 3D shape segmentation. We propose a cross‐shape attention mechanism to enable interactions between a shape's point‐wise features and those of other shapes. The mechanism assesses both the degree of interaction between points and also mediates feature propagation across shapes, improving the accuracy and consistency of the resulting point‐wise feature representations for shape segmentation. Our method also proposes a shape retrieval measure to select suitable shapes for cross‐shape attention operations for each test shape. Our experiments demonstrate that our approach yields state‐of‐the‐art results in the popular PartNet dataset.

End-to-end network performance evaluation and dynamic resource provisioning require models that are fast in execution and produce predictions in a probabilistic way, including accuracy estimations. mmWave mobile networks are challenging for the analysis due to the difference in line of sight (LoS) and non-line of sight (NLoS) regimes. The training and accuracy of the models depend on the amount of available measurement data and domain knowledge. In this paper, we consider two probabilistic models for path loss prediction in mmWave networks. Both, a Bayesian learning and a Mixture Density neural Network (MDN) models are developed and trained to predict path loss distributions in a realistic city environment based on a limited amount of training data. We measure prediction capability in terms of Kullback-Leibler (KL) divergence and Total Variation Distance (TVD). The results show that MDN describes path loss more accurately for larger training data-sets. However, the Bayesian learning predictor is more data-efficient.

Automated and timely detection of malfunctioning cells in Long-Term Evolution (LTE) networks is of high importance. Sleeping cell is a particular type of cell degradation hardly detectable by traditional network monitoring systems. Recent introduction of Minimization of Drive Test (MDT) functionality enables to collect user-level statistics from regular user devices without expensive and time-consuming drive-test and measurement campaigns. In this study data mining techniques are used to process MDT measurements to detect efficiently a sleeping cell. The developed earlier data mining framework is briefly overviewed in the paper. Special attention is devoted to post-processing stage as one of the key elements of the detection scheme. In practice, location information of collected measurements might contain considerable errors. This factor impacts the precision of malfunctioning cell detection. Therefore several post-processing algorithms are proposed, where location accuracy is taken into account. The performance of the algorithms is compared based on the results of thorough system-level LTE network simulations. Combined post-processing method shows the best reliability against location errors in terms of Root Mean Squared Error (RMSE) and percent gain.

Sleeping cell problem is a particular type of cell degradation. There are various software and hardware reasons that might cause such kind of cell outage. In this study a cell becomes sleeping because of Random Access Channel (RACH) failure. This kind of network problem can appear due to misconfiguration, excessive load or software/firmware problem at the Base Station (BS). In practice such failure might cause network performance degradation, which is hardly traceable by an operator. In this paper we present a data mining based framework for the detection of problematic cells. In its core is the analysis of event sequences reported by a User Equipment (UE) to a serving BS. The choice of N in N-gram feature selection algorithm is considered, because of its significant impact on computational efficiency. Moreover, qualitative and heuristic performance metrics have been developed to assess the performance of the proposed detection algorithm. Sleeping cell detection framework is verified by means of dynamic LTE (Long-Term Evolution) system simulator, using Minimization of Drive Testing (MDT) functionality. It is shown that sleeping cell can be determined with very high reliability even using 1-gram algorithm.

WCDMA based third generation networks are widely deployed and 3GPP is constantly seeking to evolve their performance. It is a well-known problem that reception quality degrades at cell edges. Multipoint transmission from two separate cells can be used to overcome this problem in single frequency deployments. In this paper we make an overview of several techniques for single frequency multipoint transmission, which are under consideration in 3GPP. This study presents the simulation results of Intra-site Multiflow Aggregation scheme where UEs can receive two independent transmissions from separate sectors. The performance of Type 3 and Type 3i receivers is also benchmarked in such scenario. Considerable improvement of radio network performance is demonstrated for users with Type 3i receivers on the borders of the cells without any sensible negative effect for all other users.

Mitigation of inter-carrier (ICI) and inter-symbol interference (ISI) in the systems based on Orthogonal Frequency Division Multiplexing (OFDM) has been a popular research topic already for several decades. Utilisation of signal bases with ameliorated time-frequency (TF) localisation is one of the methods to increase interference robustness. However, this approach results in more complicated signal processing. In this paper we theoretically derive and formulate computationally efficient algorithms for modulation of discrete finite-dimensional signals constructed from well-localised bases. These results can be used directly for software implementation. In addition, we present the comparison of computing load for straightforward matrix multiplication, for developed algorithm and for Fast Fourier Transform (FFT) used in OFDM. This study shows that the number of operations can be significantly reduced allowing signal processing with a great number of sub-carriers and time shifts in a reasonable time.