GenLie: A Global-Enhanced Lie Detection Network under Sparsity and Semantic Interference

Zongshun Zhang; Yao Liu; Qiao Liu; Xuefeng Peng; Peiyuan Jiang; Jiaye Yang; Daibing Yao; Wei Lin

cs.CVMar 14, 2026

GenLie: A Global-Enhanced Lie Detection Network under Sparsity and Semantic Interference

Authors: Zongshun Zhang, Yao Liu, Qiao Liu, Xuefeng Peng, Peiyuan Jiang, Jiaye Yang, Daibing Yao, Wei Lin

Organizations: School of Computer Science and Engineering, University of Electronic Science and Technology of China, Sichuan, China · Yizhou Prison, Sichuan, China

Abstract

Video-based lie detection aims to identify deceptive behaviors from visual cues. Despite recent progress, its core challenge lies in learning sparse yet discriminative representations. Deceptive signals are typically subtle and short-lived, easily overwhelmed by redundant information, while individual and contextual variations introduce strong identity-related noise. To address this issue, we propose GenLie, a Global-Enhanced Lie Detection Network that performs local feature modeling under global supervision. Specifically, sparse and subtle deceptive cues are captured at the local level, while global supervision and optimization ensure robust and discriminative representations by suppressing identity-related noise. Experiments on three public datasets, covering both high- and low-stakes scenarios, show that GenLie consistently outperforms state-of-the-art methods. Source code is available at https://github.com/AliasDictusZ1/GenLie.

Figures & tables

Figure 1: Core challenge in video-based lie detection: subtle and transient cues are often overwhelmed by noise and contextual bias.

Figure 2: Architecture of the proposed GenLie framework.

Method	MDPE			Real-Life Trial			SEUMLD
Method	F1	ACC	AUC	F1	ACC	AUC	F1	ACC	AUC
MBF [ 16 ]	0.00	60.00	48.46	0.00	49.59	0.50	13.08	64.14	51.76
MAD-Net [ 2 ]	1.02	60.00	51.70	91.80	91.74	98.80	2.10	65.32	52.44
FTM [ 7 ]	0.00	60.00	52.93	61.67	61.98	65.79	42.08	46.65	50.05
FADe-Net [ 1 ]	0.00	60.00	48.46	0.00	49.59	0.50	13.08	64.14	51.76
FFCSN [ 9 ]	37.65	57.10	53.02	90.48	90.08	97.51	41.53	62.03	56.30

Table 1: Performance comparison of baselines and GenLie on three datasets.

	MDPE			SEUMLD
Model	F1	ACC	AUC	F1	ACC	AUC
GenLie	38.97	63.01	59.91	42.39	64.17	59.15
w/o TS	24.95	60.10	53.49	38.59	57.95	51.92
w/o $\mathcal{F}_{\text{SR}}$	18.10	60.62	58.10	35.84	60.36	55.03
w/o $\mathcal{L}_{\text{id}}$	34.06	62.28	59.78	36.79	57.06	53.55
w/o $\mathcal{L}_{\text{tri}}$	25.73	62.90	58.07	40.85	61.29	56.83

Table 2: Ablation study results on MDPE and SEUMLD.

	MDPE			SEUMLD
Strategy	F1	ACC	AUC	F1	ACC	AUC
Uniform	38.97	63.01	59.91	42.39	64.17	59.15
AU-based	35.05	57.62	53.06	38.14	60.76	59.16
Gaze	26.87	60.52	55.49	42.72	58.75	57.36
Micro-exp	20.73	59.59	54.46	42.40	57.35	57.44
Posture	33.04	60.93	59.16	42.11	58.56	57.75

Table 3: Comparison of frame selection strategies on MDPE and SEUMLD.

Figure 3: Visualization of attention maps from FacialCueNet, AFFAKT, and GenLie on a deceptive clip.

Video Representation Learning

Video representation learning develops features that capture appearance, motion, and temporal structure for recognition, prediction, and control.

See all 143 papers

Catching Lies Without Sending the Video: Privacy-Preserving Multimodal Deception Detection
Jun 21, 2026Nikita Sharma, Pranav Sara, Karan SinglaPrivacy-Preserving ML Multimodal Deception Detection
University of California, San Diego · Case Western Reserve University, Ohio · WhissleAI, USA
DeceptionX: Explainable Deception Detection with Multimodal Large Language Models
Jun 9, 2026Jiayu Zhang, Shuo Ye, Jiajian Huang +7Multimodal Large Language Models Multimodal Reasoning
Great Bay University Dongguan, Guangdong, China · Hong Kong Polytechnic University HongKong, China
DecepGPT: Schema-Driven Deception Detection with Multicultural Datasets and Robust Multimodal Learning
Mar 25, 2026Jiajian Huang, Dongliang Zhu, Zitong YU +4Multimodal Robustness Cross-Domain Generalization
Great Bay University, China · Wuhan University, China · Dongguan Key Laboratory for Intelligence and Information Technology, China +1

Jun 21, 2026J/K move · Enter open · S save

Catching Lies Without Sending the Video: Privacy-Preserving Multimodal Deception Detection

Nikita Sharma, Pranav Sara, Karan Singla

University of California, San Diego · Case Western Reserve University, Ohio · WhissleAI, USA

Privacy-Preserving ML Multimodal Deception Detection Multimodal Model Evaluation

Open paper AskPDF arXiv

Frontier multimodal models can guess whether a person is lying from a testimony video. To do so, they stream that raw face and voice to a third-party model. We ask whether the heavy media is needed at all. On the Real-life Trial Deception dataset, Whissle on-device speech and vision stack extracts a compact digest: transcript, emotion, age, gender, intent distributions, a deception intent filter, fluency and rhythm, per-frame facial behaviour, and prosody. Under speaker-independent evaluation, we report three findings. A small classifier on this digest reaches AUC 0.741, matching Gemini 2.5 Pro on full video. Handing the digest to a frontier LLM reaches AUC 0.755 with Claude Opus 4.8 at 7.8X fewer input tokens, with no media leaving the device. The reported 75% accuracy is a speaker-leakage artifact. We release code and experiments.

Abstract

Video Representation Learning

Explore similar work

Catching Lies Without Sending the Video: Privacy-Preserving Multimodal Deception Detection

DeceptionX: Explainable Deception Detection with Multimodal Large Language Models

DecepGPT: Schema-Driven Deception Detection with Multicultural Datasets and Robust Multimodal Learning