Yifan Wang

Xiuwei Xu, Yifan Wang, Yu Zheng et al. · tsinghua

In this paper, we propose a weakly-supervised approach for 3D object detection, which makes it possible to train a strong 3D detector with position-level annotations (i.e. annotations of object centers). In order to remedy the information loss from box annotations to centers, our method, namely Back to Reality (BR), makes use of synthetic 3D shapes to convert the weak labels into fully-annotated virtual scenes as stronger supervision, and in turn utilizes the perfect virtual labels to complement and refine the real labels. Specifically, we first assemble 3D shapes into physically reasonable virtual scenes according to the coarse scene layout extracted from position-level annotations. Then we go back to reality by applying a virtual-to-real domain adaptation method, which refine the weak labels and additionally supervise the training of detector with the virtual scenes. Furthermore, we propose a more challenging benckmark for indoor 3D object detection with more diversity in object sizes to better show the potential of BR. With less than 5% of the labeling labor, we achieve comparable detection performance with some popular fully-supervised approaches on the widely used ScanNet dataset. Code is available at: https://github.com/wyf-ACCEPT/BackToReality

Bowei Chen, Yifan Wang, Brian Curless et al. · uw

Given an input painting, we reconstruct a time-lapse video of how it may have been painted. We formulate this as an autoregressive image generation problem, in which an initially blank "canvas" is iteratively updated. The model learns from real artists by training on many painting videos. Our approach incorporates text and region understanding to define a set of painting "instructions" and updates the canvas with a novel diffusion-based renderer. The method extrapolates beyond the limited, acrylic style paintings on which it has been trained, showing plausible results for a wide range of artistic styles and genres.

Yueqian Wang, Xiaojun Meng, Yifan Wang et al.

With the growing research focus on multimodal dialogue systems, the capability for proactive interaction is gradually gaining recognition. As an alternative to conventional turn-by-turn dialogue, users increasingly expect multimodal systems to be more initiative, for example, by autonomously determining the timing of multi-turn responses in real time during video playback. To facilitate progress in this emerging area, we introduce ProactiveVideoQA, the first comprehensive benchmark to evaluate a system's ability to engage in proactive interaction. Since model responses are generated at varying timestamps, we further propose PAUC, the first metric that accounts for the temporal dynamics of model responses. This enables a more accurate evaluation of systems operating in proactive settings. Through extensive benchmarking of various baseline systems on ProactiveVideoQA and a user study of human preferences, we show that PAUC is in better agreement with human preferences than traditional evaluation metrics, which typically only consider the textual content of responses. These findings demonstrate that PAUC provides a more faithful assessment of user experience in proactive interaction scenarios. Project homepage: https://github.com/yellow-binary-tree/ProactiveVideoQA

Qianrui Zhou, Hua Xu, Yifan Wang et al. · tsinghua

Understanding human intents from multimodal signals is critical for analyzing human behaviors and enhancing human-machine interactions in real-world scenarios. However, existing methods exhibit limitations in their modality-level reliance, constraining relational reasoning over fine-grained semantics for complex intent understanding. This paper proposes a novel LLM-Guided Semantic Relational Reasoning (LGSRR) method, which harnesses the expansive knowledge of large language models (LLMs) to establish semantic foundations that boost smaller models' relational reasoning performance. Specifically, an LLM-based strategy is proposed to extract fine-grained semantics as guidance for subsequent reasoning, driven by a shallow-to-deep Chain-of-Thought (CoT) that autonomously uncovers, describes, and ranks semantic cues by their importance without relying on manually defined priors. Besides, we formally model three fundamental types of semantic relations grounded in logical principles and analyze their nuanced interplay to enable more effective relational reasoning. Extensive experiments on multimodal intent and dialogue act recognition tasks demonstrate LGSRR's superiority over state-of-the-art methods, with consistent performance gains across diverse semantic understanding scenarios. The complete data and code are available at https://github.com/thuiar/LGSRR.

Yifan Wang, Aleksander Holynski, Brian L. Curless et al.

We present Infinite Texture, a method for generating arbitrarily large texture images from a text prompt. Our approach fine-tunes a diffusion model on a single texture, and learns to embed that statistical distribution in the output domain of the model. We seed this fine-tuning process with a sample texture patch, which can be optionally generated from a text-to-image model like DALL-E 2. At generation time, our fine-tuned diffusion model is used through a score aggregation strategy to generate output texture images of arbitrary resolution on a single GPU. We compare synthesized textures from our method to existing work in patch-based and deep learning texture synthesis methods. We also showcase two applications of our generated textures in 3D rendering and texture transfer.

Kunhao Yuan, Gerald Schaefer, Yu-Kun Lai et al.

Weakly supervised semantic segmentation (WSSS) has gained significant popularity since it relies only on weak labels such as image level annotations rather than pixel level annotations required by supervised semantic segmentation (SSS) methods. Despite drastically reduced annotation costs, typical feature representations learned from WSSS are only representative of some salient parts of objects and less reliable compared to SSS due to the weak guidance during training. In this paper, we propose a novel Multi-Strategy Contrastive Learning (MuSCLe) framework to obtain enhanced feature representations and improve WSSS performance by exploiting similarity and dissimilarity of contrastive sample pairs at image, region, pixel and object boundary levels. Extensive experiments demonstrate the effectiveness of our method and show that MuSCLe outperforms the current state-of-the-art on the widely used PASCAL VOC 2012 dataset.

Yifan Wang, Yuchen Zhang, Chengbin Pang et al.

Fuzz testing, or fuzzing, has become one of the de facto standard techniques for bug finding in the software industry. In general, fuzzing provides various inputs to the target program to discover unhandled exceptions and crashes. In business sectors where the time budget is limited, software vendors often launch many fuzzing instances in parallel as common means of increasing code coverage. However, most of the popular fuzzing tools in their parallel mode-naively run multiple instances concurrently, without elaborate distribution of workload. This can lead different instances to explore overlapped code regions, eventually reducing the benefits of concurrency. In this paper, we propose a general model to describe parallel fuzzing. This model distributes mutually-exclusive but similarly-weighted tasks to different instances, facilitating concurrency and also fairness across instances. Following this model, we develop a solution, called AFL-EDGE, to improve the parallel mode of AFL, considering a round of mutations to a unique seed as a task and adopting edge coverage to define the uniqueness of a seed. We have implemented AFL-EDGE on top of AFL and evaluated the implementation with AFL on 9 widely used benchmark programs. It shows that AFL-EDGE can benefit the edge coverage of AFL. In a 24-hour test, the increase of edge coverage brought by AFL-EDGE to AFL ranges from 9.49% to 10.20%, depending on the number of instances. As a side benefit, we discovered 14 previously unknown bugs.

Yifan Wang, Guoli Yan, Haikuan Zhu et al.

The motivation of our work is to present a new visualization-guided computing paradigm to combine direct 3D volume processing and volume rendered clues for effective 3D exploration such as extracting and visualizing microstructures in-vivo. However, it is still challenging to extract and visualize high fidelity 3D vessel structure due to its high sparseness, noisiness, and complex topology variations. In this paper, we present an end-to-end deep learning method, VC-Net, for robust extraction of 3D microvasculature through embedding the image composition, generated by maximum intensity projection (MIP), into 3D volume image learning to enhance the performance. The core novelty is to automatically leverage the volume visualization technique (MIP) to enhance the 3D data exploration at deep learning level. The MIP embedding features can enhance the local vessel signal and are adaptive to the geometric variability and scalability of vessels, which is crucial in microvascular tracking. A multi-stream convolutional neural network is proposed to learn the 3D volume and 2D MIP features respectively and then explore their inter-dependencies in a joint volume-composition embedding space by unprojecting the MIP features into 3D volume embedding space. The proposed framework can better capture small / micro vessels and improve vessel connectivity. To our knowledge, this is the first deep learning framework to construct a joint convolutional embedding space, where the computed vessel probabilities from volume rendering based 2D projection and 3D volume can be explored and integrated synergistically. Experimental results are compared with the traditional 3D vessel segmentation methods and the deep learning state-of-the-art on public and real patient (micro-)cerebrovascular image datasets. Our method demonstrates the potential in a powerful MR arteriogram and venogram diagnosis of vascular diseases.

Xingzhou Zhang, Yifan Wang, Weisong Shi

Machine learning has changed the computing paradigm. Products today are built with machine intelligence as a central attribute, and consumers are beginning to expect near-human interaction with the appliances they use. However, much of the deep learning revolution has been limited to the cloud. Recently, several machine learning packages based on edge devices have been announced which aim to offload the computing to the edges. However, little research has been done to evaluate these packages on the edges, making it difficult for end users to select an appropriate pair of software and hardware. In this paper, we make a performance comparison of several state-of-the-art machine learning packages on the edges, including TensorFlow, Caffe2, MXNet, PyTorch, and TensorFlow Lite. We focus on evaluating the latency, memory footprint, and energy of these tools with two popular types of neural networks on different edge devices. This evaluation not only provides a reference to select appropriate combinations of hardware and software packages for end users but also points out possible future directions to optimize packages for developers.