Yichen Yan

Yichen Yan, Xingjian He, Wenxuan Wang et al.

Referring image segmentation (RIS) aims to segment an object mentioned in natural language from an image. The main challenge is text-to-pixel fine-grained correlation. In the previous methods, the final results are obtained by convolutions with a fixed kernel, which follows a similar pattern as traditional image segmentation. These methods lack explicit alignment of language and vision features in the segmentation stage, resulting in suboptimal correlation. In this paper, we introduce EAVL, a method explicitly aligning vision and language features. In contrast to fixed convolution kernels, we introduce a Vision-Language Aligner that aligns features in the segmentation stage using dynamic convolution kernels based on the input image and sentence. Specifically, we generate multiple queries representing different emphases of language expression. These queries are transformed into a series of query-based convolution kernels, which are applied in the segmentation stage to produce a series of masks. The final result is obtained by aggregating all masks. Our method harnesses the potential of the multi-modal features in the segmentation stage and aligns language features of different emphases with image features to achieve fine-grained text-to-pixel correlation. We surpass previous state-of-the-art methods on RefCOCO, RefCOCO+, and G-Ref by large margins. Additionally, our method is designed to be a generic plug-and-play module for cross-modality alignment in RIS task, making it easy to integrate with other RIS models for substantial performance improvements.

Wenxuan Wang, Yisi Zhang, Xingjian He et al.

Visual grounding (VG) aims at locating the foreground entities that match the given natural language expressions. Previous datasets and methods for classic VG task mainly rely on the prior assumption that the given expression must literally refer to the target object, which greatly impedes the practical deployment of agents in real-world scenarios. Since users usually prefer to provide intention-based expression for the desired object instead of covering all the details, it is necessary for the agents to interpret the intention-driven instructions. Thus, in this work, we take a step further to the intention-driven visual-language (V-L) understanding. To promote classic VG towards human intention interpretation, we propose a new intention-driven visual grounding (IVG) task and build a large-scale IVG dataset termed IntentionVG with free-form intention expressions. Considering that practical agents need to move and find specific targets among various scenarios to realize the grounding task, our IVG task and IntentionVG dataset have taken the crucial properties of both multi-scenario perception and egocentric view into consideration. Besides, various types of models are set up as the baselines to realize our IVG task. Extensive experiments on our IntentionVG dataset and baselines demonstrate the necessity and efficacy of our method for the V-L field. To foster future research in this direction, our newly built dataset and baselines will be publicly available at https://github.com/Rubics-Xuan/IVG.

Ting Su, Chengyu Zhang, Yichen Yan et al.

Data-flow testing (DFT) aims to detect potential data interaction anomalies by focusing on the points at which variables receive values and the points at which these values are used. Such test objectives are referred as \emph{def-use pairs}. However, the complexity of DFT still overwhelms the testers in practice. To tackle this problem, we introduce a hybrid testing framework for data-flow based test generation: (1) The core of our framework is symbolic execution (SE), enhanced by a novel guided path exploration strategy to improve testing performance; and (2) we systematically cast DFT as reachability checking in software model checking (SMC) to complement SE, yielding practical DFT that combines the two techniques' strengths. We implemented our framework for C programs on top of the state-of-the-art symbolic execution engine KLEE and instantiated with three different software model checkers. Our evaluation on the 28,354 def-use pairs collected from 33 open-source and industrial program subjects shows (1) our SE-based approach can improve DFT performance by 15$\sim$48% in terms of testing time, compared with existing search strategies; and (2) our combined approach can further reduce testing time by 20.1$\sim$93.6%, and improve data-flow coverage by 27.8$\sim$45.2% by eliminating infeasible test objectives. Compared with the SMC-based approach alone, our combined approach can also reduce testing time by 19.9$\sim$23.8%, and improve data-flow coverage by 7$\sim$10%. This combined approach also enables the cross-checking of each component for reliable and robust testing results. We have made our testing framework and benchmarks publicly available to facilitate future research.

Chengyu Zhang, Yichen Yan, Hanru Zhou et al.

In this paper, we aim at the automated unit coverage-based testing for embedded software. To achieve the goal, by analyzing the industrial requirements and our previous work on automated unit testing tool CAUT, we rebuild a new tool, SmartUnit, to solve the engineering requirements that take place in our partner companies. SmartUnit is a dynamic symbolic execution implementation, which supports statement, branch, boundary value and MC/DC coverage. SmartUnit has been used to test more than one million lines of code in real projects. For confidentiality motives, we select three in-house real projects for the empirical evaluations. We also carry out our evaluations on two open source database projects, SQLite and PostgreSQL, to test the scalability of our tool since the scale of the embedded software project is mostly not large, 5K-50K lines of code on average. From our experimental results, in general, more than 90% of functions in commercial embedded software achieve 100% statement, branch, MC/DC coverage, more than 80% of functions in SQLite achieve 100% MC/DC coverage, and more than 60% of functions in PostgreSQL achieve 100% MC/DC coverage. Moreover, SmartUnit is able to find the runtime exceptions at the unit testing level. We also have reported exceptions like array index out of bounds and divided-by-zero in SQLite. Furthermore, we analyze the reasons of low coverage in automated unit testing in our setting and give a survey on the situation of manual unit testing with respect to automated unit testing in industry.

Yichen Yan, Xingjian He, Sihan Chen et al.

Referring image segmentation aims to segment an object referred to by natural language expression from an image. The primary challenge lies in the efficient propagation of fine-grained semantic information from textual features to visual features. Many recent works utilize a Transformer to address this challenge. However, conventional transformer decoders can distort linguistic information with deeper layers, leading to suboptimal results. In this paper, we introduce CRFormer, a model that iteratively calibrates multi-modal features in the transformer decoder. We start by generating language queries using vision features, emphasizing different aspects of the input language. Then, we propose a novel Calibration Decoder (CDec) wherein the multi-modal features can iteratively calibrated by the input language features. In the Calibration Decoder, we use the output of each decoder layer and the original language features to generate new queries for continuous calibration, which gradually updates the language features. Based on CDec, we introduce a Language Reconstruction Module and a reconstruction loss. This module leverages queries from the final layer of the decoder to reconstruct the input language and compute the reconstruction loss. This can further prevent the language information from being lost or distorted. Our experiments consistently show the superior performance of our approach across RefCOCO, RefCOCO+, and G-Ref datasets compared to state-of-the-art methods.

Yichen Yan, Xingjian He, Sihan Chen et al.

Referring Image Segmentation (RIS) aims to segment an object described in natural language from an image, with the main challenge being a text-to-pixel correlation. Previous methods typically rely on single-modality features, such as vision or language features, to guide the multi-modal fusion process. However, this approach limits the interaction between vision and language, leading to a lack of fine-grained correlation between the language description and pixel-level details during the decoding process. In this paper, we introduce FCNet, a framework that employs a bi-directional guided fusion approach where both vision and language play guiding roles. Specifically, we use a vision-guided approach to conduct initial multi-modal fusion, obtaining multi-modal features that focus on key vision information. We then propose a language-guided calibration module to further calibrate these multi-modal features, ensuring they understand the context of the input sentence. This bi-directional vision-language guided approach produces higher-quality multi-modal features sent to the decoder, facilitating adaptive propagation of fine-grained semantic information from textual features to visual features. Experiments on RefCOCO, RefCOCO+, and G-Ref datasets with various backbones consistently show our approach outperforming state-of-the-art methods.

Yichen Yan, Ming Zhong, Qi Zhu et al.

Multimodal large language models (MLLMs) rely heavily on instruction tuning to align vision and language capabilities, yet the computational cost of training on large-scale datasets remains a major bottleneck. Existing data selection methods aim to mitigate this by selecting important and diverse subsets, but they often suffer from two critical drawbacks: high computational overhead from processing the entire dataset and suboptimal data selection due to separate treatment of importance and diversity. We introduce CoIDO, a novel dual-objective framework that jointly optimizes data importance and diversity to overcome these challenges. Unlike existing approaches that require costly evaluations across the whole dataset, CoIDO employs a lightweight plug-in scorer. This scorer is trained on just a small random sample of data to learn the distribution of the candidate set, drastically reducing computational demands. By leveraging a homoscedastic uncertainty-based formulation, CoIDO effectively balances importance and diversity during training, enabling efficient and scalable data selection. In our experiments, we trained the CoIDO scorer using only 20 percent of randomly sampled data. Once trained, CoIDO was applied to the entire dataset to select a 20 percent subset for instruction tuning. On the widely used LLaVA-1.5-7B model across ten downstream tasks, this selected subset achieved an impressive 98.2 percent of the performance of full-data fine-tuning, on average.

Jingen Qu, Lijun Li, Bo Zhang et al.

Multimodal large language models (MLLMs) are rapidly evolving, presenting increasingly complex safety challenges. However, current dataset construction methods, which are risk-oriented, fail to cover the growing complexity of real-world multimodal safety scenarios (RMS). And due to the lack of a unified evaluation metric, their overall effectiveness remains unproven. This paper introduces a novel image-oriented self-adaptive dataset construction method for RMS, which starts with images and end constructing paired text and guidance responses. Using the image-oriented method, we automatically generate an RMS dataset comprising 35k image-text pairs with guidance responses. Additionally, we introduce a standardized safety dataset evaluation metric: fine-tuning a safety judge model and evaluating its capabilities on other safety datasets.Extensive experiments on various tasks demonstrate the effectiveness of the proposed image-oriented pipeline. The results confirm the scalability and effectiveness of the image-oriented approach, offering a new perspective for the construction of real-world multimodal safety datasets.

Yichen Yan, Xingjian He, Wenxuan Wan et al.

Referring image segmentation aims to segment an object referred to by natural language expression from an image. However, this task is challenging due to the distinct data properties between text and image, and the randomness introduced by diverse objects and unrestricted language expression. Most of previous work focus on improving cross-modal feature fusion while not fully addressing the inherent uncertainty caused by diverse objects and unrestricted language. To tackle these problems, we propose an end-to-end Multi-Mask Network for referring image segmentation(MMNet). we first combine picture and language and then employ an attention mechanism to generate multiple queries that represent different aspects of the language expression. We then utilize these queries to produce a series of corresponding segmentation masks, assigning a score to each mask that reflects its importance. The final result is obtained through the weighted sum of all masks, which greatly reduces the randomness of the language expression. Our proposed framework demonstrates superior performance compared to state-of-the-art approaches on the two most commonly used datasets, RefCOCO, RefCOCO+ and G-Ref, without the need for any post-processing. This further validates the efficacy of our proposed framework.

Ting Su, Yichen Yan, Jue Wang et al.

Android apps are GUI-based event-driven software and have become ubiquitous in recent years. Obviously, functional correctness is critical for an app's success. However, in addition to crash bugs, non-crashing functional bugs (in short as "non-crashing bugs" in this work) like inadvertent function failures, silent user data lost and incorrect display information are prevalent, even in popular, well-tested apps. These non-crashing functional bugs are usually caused by program logic errors and manifest themselves on the graphic user interfaces (GUIs). In practice, such bugs pose significant challenges in effectively detecting them because (1) current practices heavily rely on expensive, small-scale manual validation (the lack of automation); and (2) modern fully automated testing has been limited to crash bugs (the lack of test oracles). This paper fills this gap by introducing independent view fuzzing, a novel, fully automated approach for detecting non-crashing functional bugs in Android apps. Inspired by metamorphic testing, our key insight is to leverage the commonly-held independent view property of Android apps to manufacture property-preserving mutant tests from a set of seed tests that validate certain app properties. The mutated tests help exercise the tested apps under additional, adverse conditions. Any property violations indicate likely functional bugs for further manual confirmation. We have realized our approach as an automated, end-to-end functional fuzzing tool, Genie. Given an app, (1) Genie automatically detects non-crashing bugs without requiring human-provided tests and oracles (thus fully automated); and (2) the detected non-crashing bugs are diverse (thus general and not limited to specific functional properties), which set Genie apart from prior work.

Chengyu Zhang, Ting Su, Yichen Yan et al.

Dataflow coverage, one of the white-box testing criteria, focuses on the relations between variable definitions and their uses.Several empirical studies have proved data-flow testing is more effective than control-flow testing. However, data-flow testing still cannot find its adoption in practice, due to the lack of effective tool support. To this end, we propose a guided symbolic execution approach to efficiently search for program paths to satisfy data-flow coverage criteria. We implemented this approach on KLEE and evaluated with 30 program subjects which are constructed by the subjects used in previous data-flow testing literature, SIR, SV-COMP benchmarks. Moreover, we are planning to integrate the data-flow testing technique into the new proposed symbolic execution engine, SmartUnit, which is a cloud-based unit testing service for industrial software, supporting coverage-based testing. It has successfully helped several well-known corporations and institutions in China to adopt coverage-based testing in practice, totally tested more than one million lines of real code from industry.