Generalizable 3D Gaussian Splatting reconstruction showcases advanced Image-to-3D content creation but requires substantial computational resources and large datasets, posing challenges to training models from scratch. Current methods usually entangle the prediction of 3D Gaussian geometry and appearance, which rely heavily on data-driven priors and result in slow regression speeds. To address this, we propose Stereo-GS, a disentangled framework for efficient 3D Gaussian prediction. Our method extracts features from local image pairs using a stereo vision backbone and fuses them via global attention blocks. Dedicated point and Gaussian prediction heads generate multi-view point-maps for geometry and Gaussian features for appearance, combined as GS-maps to represent the 3DGS object. A refinement network enhances these GSmaps for high-quality reconstruction. Unlike existing methods that depend on camera parameters, our approach achieves pose-free 3D reconstruction, improving robustness and practicality. By reducing resource demands while maintaining high-quality outputs, Stereo- GS provides an efficient, scalable solution for real-world 3D content generation. Project page: https://kevinhuangxf.github.io/stereo-gs.

The growing popularity of 3D Gaussian Splatting (3DGS) has intensified the need for effective copyright protection. Current 3DGS watermarking methods rely on computationally expensive fine-tuning procedures for each predefined message. We propose the first generalizable watermarking framework that enables efficient protection of Splatter Image-based 3DGS models through a single forward pass. We introduce GaussianBridge that transforms unstructured 3D Gaussians into Splatter Image format, enabling direct neural processing for arbitrary message embedding. To ensure imperceptibility, we design a Gaussian-Uncertainty-Perceptual heatmap prediction strategy for preserving visual quality. For robust message recovery, we develop a dense segmentation-based extraction mechanism that maintains reliable extraction even when watermarked objects occupy minimal regions in rendered views. Project page: https://kevinhuangxf.github.io/marksplatter.

3D Gaussian Splatting (3DGS) has become a crucial method for acquiring 3D assets. To protect the copyright of these assets, digital watermarking techniques can be applied to embed ownership information discreetly within 3DGS mod- els. However, existing watermarking methods for meshes, point clouds, and implicit radiance fields cannot be directly applied to 3DGS models, as 3DGS models use explicit 3D Gaussians with distinct structures and do not rely on neural networks. Naively embedding the watermark on a pre-trained 3DGS can cause obvious distortion in rendered images. In our work, we propose an uncertainty- based method that constrains the perturbation of model parameters to achieve invisible watermarking for 3DGS. At the message decoding stage, the copyright messages can be reliably extracted from both 3D Gaussians and 2D rendered im- ages even under various forms of 3D and 2D distortions. We conduct extensive experiments on the Blender, LLFF, and MipNeRF-360 datasets to validate the effectiveness of our proposed method, demonstrating state-of-the-art performance on both message decoding accuracy and view synthesis quality.

To improve the performance of multi-agent reinforcement learning under the constraint of wireless resources, we propose a message importance metric and design an importance-aware scheduling policy to effectively exchange messages. The key insight is spending the precious communication resources on important messages. The message importance depends not only on the messages themselves, but also on the needs of agents who receive them. Accordingly, we propose a query-message-based architecture, called QMNet. Agents generate queries and messages with the environment observation. Sharing queries can help calculate message importance. Exchanging messages can help agents cooperate better. Besides, we exploit the message importance to deal with random access collisions in decentralized systems. Furthermore, a message prediction mechanism is proposed to compensate for messages that are not transmitted. Finally, we evaluate the proposed schemes in a traffic junction environment, where only a fraction of agents can send messages due to limited wireless resources. Results show that QMNet can extract valuable information to guarantee the system performance even when only 30% of agents can share messages. By exploiting message prediction, the system can further save 40% of wireless resources. The importance-aware decentralized multi-access mechanism can effectively avoid collisions, achieving almost the same performance as centralized scheduling.

The cooperation among intelligent agents is the key to build smarter real-world intelligent systems. Recent researches have shown that wireless communication plays a vital role in multi-agent cooperation. However, limited wireless resources become the bottleneck of large-scale multi-agent cooperation. Therefore, we focus on how to improve the performance of multi-agent reinforcement learning under the constraint of communication resources. To share more valuable information among agents under resource constraint, we formulate the message importance and design a decentralized scheduling policy with query mechanism, so that agents can effectively exchange messages according to their message importance. We further design a message prediction mechanism to compensate for those messages that are not scheduled for transmission in the current round. Finally, we evaluate the performance of the proposed schemes in a traffic junction environment, where only a fraction of agents can broadcast their messages in each round due to limited wireless resources. Results show that the importance-aware message exchange can extract valuable information to guarantee the system performance even when less than half of agents can share their states. By exploiting message prediction, the system can further save 40% of communication resources while guaranteeing the system performance.

With the large-scale deployment of intelligent Internet of things (IoT) devices and the increasing need for computation support in wireless access networks, edge computing plays a vital role in satisfying these merging needs. The deployment of machine learning algorithms as one of the key applications at the network edge requires efficient training, in order to adapt themselves in the changing environment. However, the transmission of the training dataset collected by edge devices requires huge wireless communication resources. To address this issue, we exploit the fact that data samples have different importance for training, and use the training loss of the target machine learning model to represent the importance of data samples. Based on the importance metric, we propose a data upload scheme combining data compression that removes unimportant information and data filtering for data selection. As a result, the number of data samples as well as the size of every data sample to be transmitted can be substantially reduced while keeping the good training performance. Experiments show that training a machine learning model via the proposed scheme can enjoy faster convergence under limited wireless resources, specifically, we get almost the same training performance with only 2.5% of communication resources in the experiments training on MNIST dataset.

With the development of intelligent Internet of things (IoT) devices, implementing machine learning algorithms at the network edge has become essential to many applications, such as autonomous driving, environment monitoring. However, the limited computation capability and energy constraint results in difficulties of running complex machine learning algorithms on edge devices subject to latency requirements, and one solution is to offload the computation tasks to the edge server. However, the wireless transmission of raw data from devices to the server is time consuming and may violate the latency requirement. To this end, lossy data compression can be helpful, but the information loss may lead to erroneous learning result, e.g., wrong classification. In this paper, we propose a transmission scheme with compression ratio selection for inference tasks with task completion latency guarantee. By dynamically selecting the optimal compression ratio with the awareness of the remaining latency budget, more tasks can be timely completed and get the correct inference results under the communication resource constraint. Furthermore, retransmitting less compressed data of tasks with erroneous inference results can potentially enhance the average accuracy. However, it is often hard to know whether the inference result is correct or not. We therefore use uncertainty to estimate the confidence of the results, and based on that, jointly optimize the retransmission and compression ratio selection.

Implementing machine learning algorithms on the Internet-of-Things (IoT) devices has become essential for emerging applications, such as autonomous driving and environment monitoring. But the limitations of computation capability and energy consumption make it difficult to run complex machine learning algorithms on IoT devices, especially when the latency deadline exists. One solution is to offload the computation-intensive tasks to the edge server. However, the wireless uploading of the raw data is time consuming and may lead to deadline violation. To reduce the communication cost, lossy data compression can be exploited for inference tasks but may bring more erroneous inference results. In this article, we propose a dynamic compression ratio selection scheme for edge inference system with hard deadlines. The key idea is to balance the tradeoff between the communication cost and inference accuracy. By dynamically selecting the optimal compression ratio with the remaining deadline budgets for queued tasks, more tasks can be timely completed with correct inference under limited communication resources. Furthermore, information augmentation that retransmits less compressed data of task with erroneous inference, is proposed to enhance the accuracy performance. While it is often hard to know the correctness of inference, we use uncertainty to estimate the confidence of the inference and based on that, jointly optimize the information augmentation and compression ratio selection. Finally, considering the wireless transmission errors, we further design a retransmission scheme to reduce performance degradation due to packet losses. The simulation results show the performance of the proposed schemes under different deadlines and task arrival rates.

Imagine taking a selfie video with your mobile phone and getting as output a 3D model of your head (face and 3D hair strands) that can be later used in VR, AR, and any other domain. State of the art hair reconstruction methods allow either a single photo (thus compromising 3D quality) or multiple views, but they require manual user interaction (manual hair segmentation and capture of fixed camera views that span full 360°). In this paper, we describe a system that can completely automatically create a reconstruction from any video (even a selfie video), and we don't require specific views, since taking your -90°, 90°, and full back views is not feasible in a selfie capture. In the core of our system, in addition to the automatization components, hair strands are estimated and deformed in 3D (rather than 2D as in state of the art) thus enabling superior results. We provide qualitative, quantitative, and Mechanical Turk human studies that support the proposed system, and show results on a diverse variety of videos (8 different celebrity videos, 9 selfie mobile videos, spanning age, gender, hair length, type, and styling).