Author name cluster

Li Cui

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

8 papers

2 author rows

IROS Conference 2024 Conference Paper

NF-SLAM: Effective, Normalizing Flow-supported Neural Field representations for object-level visual SLAM in automotive applications

Li Cui
Yang Ding
Richard I. Hartley
Zirui Xie
Laurent Kneip
Zhenghua Yu

We propose a novel, vision-only object-level SLAM framework for automotive applications representing 3D shapes by implicit signed distance functions. Our key innovation consists of augmenting the standard neural representation by a normalizing flow network. As a result, achieving strong representation power on the specific class of road vehicles is made possible by compact networks with only 16-dimensional latent codes. Furthermore, the newly proposed architecture exhibits a significant performance improvement in the presence of only sparse and noisy data, which is demonstrated through comparative experiments on synthetic data. The module is embedded into the back-end of a stereo-vision based framework for joint, incremental shape optimization. The loss function is given by a combination of a sparse 3D point-based SDF loss, a sparse rendering loss, and a semantic mask-based silhouette-consistency term. We furthermore leverage semantic information to determine keypoint extraction density in the front-end. Finally, experimental results on real-world data reveal accurate and reliable performance comparable to alternative frameworks that make use of direct depth readings. The proposed method performs well with only sparse 3D points obtained from bundle adjustment, and eventually continues to deliver stable results even under exclusive use of the mask-consistency term.

Details

AAAI Conference 2022 Conference Paper

Width & Depth Pruning for Vision Transformers

Fang Yu
Kun Huang
Meng Wang
Yuan Cheng
Wei Chu
Li Cui

Transformer models have demonstrated their promising potential and achieved excellent performance on a series of computer vision tasks. However, the huge computational cost of vision transformers hinders their deployment and application to edge devices. Recent works have proposed to find and remove the unimportant units of vision transformers. Despite achieving remarkable results, these methods take one dimension of network width into consideration and ignore network depth, which is another important dimension for pruning vision transformers. Therefore, we propose a Width & Depth Pruning (WDPruning) framework that reduces both width and depth dimensions simultaneously. Specifically, for width pruning, a set of learnable pruning-related parameters is used to adaptively adjust the width of transformer. For depth pruning, we introduce several shallow classifiers by using the intermediate information of the transformer blocks, which allows images to be classified by shallow classifiers instead of the deeper classifiers. In the inference period, all of the blocks after shallow classifiers can be dropped so they don’t bring additional parameters and computation. Experimental results on benchmark datasets demonstrate that the proposed method can significantly reduce the computational costs of mainstream vision transformers such as DeiT and Swin Transformer with a minor accuracy drop. In particular, on ILSVRC-12, we achieve over 22% pruning ratio of FLOPs by compressing DeiT-Base, even with an increase of 0. 14% Top-1 accuracy.

PDF Details

IROS Conference 2021 Conference Paper

Accurate depth estimation from a hybrid event-RGB stereo setup

Yi-Fan Zuo
Li Cui
Xin Peng 0005
Yanyu Xu 0001
Shenghua Gao
Xia Wang 0002
Laurent Kneip

Event-based visual perception is becoming increasingly popular owing to interesting sensor characteristics enabling the handling of difficult conditions such as highly dynamic motion or challenging illumination. The mostly complementary nature of event cameras however still means that best results are achieved if the sensor is paired with a regular frame-based sensor. The present work aims at answering a simple question: Assuming that both cameras do not share a common optical center, is it possible to exploit the hybrid stereo setup's baseline to perform accurate stereo depth estimation? We present a learning based solution to this problem leveraging modern spatio-temporal input representations as well as a novel hybrid pyramid attention module. Results on real data demonstrate competitive performance against pure frame-based stereo alternatives as well as the ability to maintain the advantageous properties of event-based sensors.

Details

IROS Conference 2021 Conference Paper

Monte-Carlo Localization in Underground Parking Lots using Parking Slot Numbers

Li Cui
Chunyan Rong
Jingyi Huang
Andre Rosendo
Laurent Kneip

Autonomous Valet Parking (AVP) in an under- ground garage is an emerging smart vehicle solution that the community believes to be solvable with close-to-market sensors. Absence of GPS signals and a high degree of self-similarity however render global visual localization in such environments a highly challenging problem. We present a novel underground parking localization method that relies on text recognition in the wild as well as optical character recognition (OCR) to automatically detect parking slot numbers. The detected numbers are then correlated with both geometric as well as semantic information extracted from an offline map of the environment. The resulting measurement model is embedded into a probabilistic Monte-Carlo localization framework. The success of our method is demonstrated on multiple real-world sequences in one of the largest underground parking garages in Shanghai.

Details

ECAI Conference 2020 Conference Paper

Tutor-Instructing Global Pruning for Accelerating Convolutional Neural Networks

Fang Yu 0004
Li Cui

Model compression and acceleration has recently received ever-increasing research attention. Among them, filter pruning shows a promising effectiveness, due to its merits in significant speedup for inference and support on off-the-shelf computing platforms. Most existing works tend to prune filters in a layer-wise manner, where networks are pruned and fine-tuned layer by layer. However, these methods require intensive computation for per-layer sensitivity analysis and suffer from accumulation of pruning errors. To address these challenges, we propose a novel pruning method, namely Tutor-Instructing global Pruning (TIP), to prune the redundant filters in a global manner. TIP introduces Information Gain (IG) to estimate the contribution of filters to the class probability distributions of network output. The motivation of TIP is to formulate filter pruning as a minimization of the IG with respect to a group of pruned filters under a constraint on the size of pruned network. To solve this problem, we propose a Taylor-based approximate algorithm, which can efficiently obtain the IG of each filter by backpropagation. We comprehensively evaluate our TIP on CIFAR-10 and ILSVRC-12. On ILSVRC-12, TIP reduces FLOPs for ResNet-50 by 54. 13% with only a drop in top-5 accuracy by 0. 1%, which significantly outperforms the state-of-the-art methods.

Details

IJCAI Conference 2019 Conference Paper

Relation Extraction Using Supervision from Topic Knowledge of Relation Labels

Haiyun Jiang
Li Cui
Zhe Xu
Deqing Yang
Jindong Chen
Chenguang Li
Jingping Liu
Jiaqing Liang

Explicitly exploring the semantics of a relation is significant for high-accuracy relation extraction, which is, however, not fully studied in previous work. In this paper, we mine the topic knowledge of a relation to explicitly represent the semantics of this relation, and model relation extraction as a matching problem. That is, the matching score between a sentence and a candidate relation is predicted for an entity pair. To this end, we propose a deep matching network to precisely model the semantic similarity between a sentence-relation pair. Besides, the topic knowledge also allows us to derive the importance information of samples as well as two knowledge-guided negative sampling strategies in the training process. We conduct extensive experiments to evaluate the proposed framework and observe improvements in AUC of 11. 5% and max F1 of 5. 4% over the baselines with state-of-the-art performance.

PDF Details

YNIMG Journal 2012 Journal Article

An automatic MEG low-frequency source imaging approach for detecting injuries in mild and moderate TBI patients with blast and non-blast causes

Ming-Xiong Huang
Sharon Nichols
Ashley Robb
Annemarie Angeles
Angela Drake
Martin Holland
Sarah Asmussen
John D'Andrea

Traumatic brain injury (TBI) is a leading cause of sustained impairment in military and civilian populations. However, mild (and some moderate) TBI can be difficult to diagnose because the injuries are often not detectable on conventional MRI or CT. Injured brain tissues in TBI patients generate abnormal low-frequency magnetic activity (ALFMA, peaked at 1–4Hz) that can be measured and localized by magnetoencephalography (MEG). We developed a new automated MEG low-frequency source imaging method and applied this method in 45 mild TBI (23 from combat-related blasts, and 22 from non-blast causes) and 10 moderate TBI patients (non-blast causes). Seventeen of the patients with mild TBI from blasts had tertiary injuries resulting from the blast. The results show our method detected abnormalities at the rates of 87% for the mild TBI group (blast-induced plus non-blast causes) and 100% for the moderate group. Among the mild TBI patients, the rates of abnormalities were 96% and 77% for the blast and non-blast TBI groups, respectively. The spatial characteristics of abnormal slow-wave generation measured by Z scores in the mild blast TBI group significantly correlated with those in non-blast mild TBI group. Among 96 cortical regions, the likelihood of abnormal slow-wave generation was less in the mild TBI patients with blast than in the mild non-blast TBI patients, suggesting possible protective effects due to the military helmet and armor. Finally, the number of cortical regions that generated abnormal slow-waves correlated significantly with the total post-concussive symptom scores in TBI patients. This study provides a foundation for using MEG low-frequency source imaging to support the clinical diagnosis of TBI.

Details DOI

YNIMG Journal 2007 Journal Article

A novel integrated MEG and EEG analysis method for dipolar sources

Ming-Xiong Huang
Tao Song
Donald J. Hagler
Igor Podgorny
Veikko Jousmaki
Li Cui
Kathleen Gaa
Deborah L. Harrington

The ability of magnetoencephalography (MEG) to accurately localize neuronal currents and obtain tangential components of the source is largely due to MEG's insensitivity to the conductivity profile of the head tissues. However, MEG cannot reliably detect the radial component of the neuronal current. In contrast, the localization accuracy of electroencephalography (EEG) is not as good as MEG, but EEG can detect both the tangential and radial components of the source. In the present study, we investigated the conductivity dependence in a new approach that combines MEG and EEG to accurately obtain, not only the location and tangential components, but also the radial component of the source. In this approach, the source location and tangential components are obtained from MEG alone, and optimal conductivity values of the EEG model are estimated by best-fitting EEG signal, while precisely matching the tangential components of the source in EEG and MEG. Then, the radial components are obtained from EEG using the previously estimated optimal conductivity values. Computer simulations testing this integrated approach demonstrated two main findings. First, there are well-organized optimal combinations of the conductivity values that provide an accurate fit to the combined MEG and EEG data. Second, the radial component, in addition to the location and tangential components, can be obtained with high accuracy without needing to know the precise conductivity profile of the head. We then demonstrated that this new approach performed reliably in an analysis of the 20-ms component from human somatosensory responses elicited by electric median-nerve stimulation.

Details DOI