arXiv subjects

Xiao Sun

Publications and source records attributed to Xiao Sun.

At least 109 records · Page 6Linked to original sources

arXiv2023

Bi-ViT: Pushing the Limit of Vision Transformer Quantization

Vision transformers (ViTs) quantization offers a promising prospect to facilitate deploying large pre-trained networks on resource-limited devices. Fully-binarized ViTs (Bi-ViT) that pushes the quantization of ViTs to its limit remain largely unexplored and a very challenging task yet, due to their unacceptable performance. Through extensive empirical analyses, we identify the severe drop in ViT binarization is caused by attention distortion in self-attention, which technically stems from the gradient vanishing and ranking disorder. To address these issues, we first introduce a learnable scaling factor to reactivate the vanished gradients and illustrate its effectiveness through theoretical and experimental analyses. We then propose a ranking-aware distillation method to rectify the disordered ranking in a teacher-student framework. Bi-ViT achieves significant improvements over popular DeiT and Swin backbones in terms of Top-1 accuracy and FLOPs. For example, with DeiT-Tiny and Swin-Tiny, our method significantly outperforms baselines by 22.1% and 21.4% respectively, while 61.5x and 56.1x theoretical acceleration in terms of FLOPs compared with real-valued counterparts on ImageNet.

cs.CV↗

arXiv2023

Enhancing Personalized Ranking With Differentiable Group AUC Optimization

AUC is a common metric for evaluating the performance of a classifier. However, most classifiers are trained with cross entropy, and it does not optimize the AUC metric directly, which leaves a gap between the training and evaluation stage. In this paper, we propose the PDAOM loss, a Personalized and Differentiable AUC Optimization method with Maximum violation, which can be directly applied when training a binary classifier and optimized with gradient-based methods. Specifically, we construct the pairwise exponential loss with difficult pair of positive and negative samples within sub-batches grouped by user ID, aiming to guide the classifier to pay attention to the relation between hard-distinguished pairs of opposite samples from the perspective of independent users. Compared to the origin form of pairwise exponential loss, the proposed PDAOM loss not only improves the AUC and GAUC metrics in the offline evaluation, but also reduces the computation complexity of the training objective. Furthermore, online evaluation of the PDAOM loss on the 'Guess What You Like' feed recommendation application in Meituan manifests 1.40% increase in click count and 0.65% increase in order count compared to the baseline model, which is a significant improvement in this well-developed online life service recommendation system.

cs.LG↗

arXiv2023

A Simple Multi-Modality Transfer Learning Baseline for Sign Language Translation

This paper proposes a simple transfer learning baseline for sign language translation. Existing sign language datasets (e.g. PHOENIX-2014T, CSL-Daily) contain only about 10K-20K pairs of sign videos, gloss annotations and texts, which are an order of magnitude smaller than typical parallel data for training spoken language translation models. Data is thus a bottleneck for training effective sign language translation models. To mitigate this problem, we propose to progressively pretrain the model from general-domain datasets that include a large amount of external supervision to within-domain datasets. Concretely, we pretrain the sign-to-gloss visual network on the general domain of human actions and the within-domain of a sign-to-gloss dataset, and pretrain the gloss-to-text translation network on the general domain of a multilingual corpus and the within-domain of a gloss-to-text corpus. The joint model is fine-tuned with an additional module named the visual-language mapper that connects the two networks. This simple baseline surpasses the previous state-of-the-art results on two sign language translation benchmarks, demonstrating the effectiveness of transfer learning. With its simplicity and strong performance, this approach can serve as a solid baseline for future research. Code and models are available at: https://github.com/FangyunWei/SLRT.

cs.CV↗

arXiv2023

Facial Affect Recognition based on Transformer Encoder and Audiovisual Fusion for the ABAW5 Challenge

In this paper, we present our solutions for the 5th Workshop and Competition on Affective Behavior Analysis in-the-wild (ABAW), which includes four sub-challenges of Valence-Arousal (VA) Estimation, Expression (Expr) Classification, Action Unit (AU) Detection and Emotional Reaction Intensity (ERI) Estimation. The 5th ABAW competition focuses on facial affect recognition utilizing different modalities and datasets. In our work, we extract powerful audio and visual features using a large number of sota models. These features are fused by Transformer Encoder and TEMMA. Besides, to avoid the possible impact of large dimensional differences between various features, we design an Affine Module to align different features to the same dimension. Extensive experiments demonstrate that the superiority of the proposed method. For the VA Estimation sub-challenge, our method obtains the mean Concordance Correlation Coefficient (CCC) of 0.6066. For the Expression Classification sub-challenge, the average F1 Score is 0.4055. For the AU Detection sub-challenge, the average F1 Score is 0.5296. For the Emotional Reaction Intensity Estimation sub-challenge, the average pearson's correlations coefficient on the validation set is 0.3968. All of the results of four sub-challenges outperform the baseline with a large margin.

cs.CV↗

arXiv2023

Spatial-temporal Transformer for Affective Behavior Analysis

The in-the-wild affective behavior analysis has been an important study. In this paper, we submit our solutions for the 5th Workshop and Competition on Affective Behavior Analysis in-the-wild (ABAW), which includes V-A Estimation, Facial Expression Classification and AU Detection Sub-challenges. We propose a Transformer Encoder with Multi-Head Attention framework to learn the distribution of both the spatial and temporal features. Besides, there are virious effective data augmentation strategies employed to alleviate the problems of sample imbalance during model training. The results fully demonstrate the effectiveness of our proposed model based on the Aff-Wild2 dataset.

cs.CV↗

arXiv2023

Mutilmodal Feature Extraction and Attention-based Fusion for Emotion Estimation in Videos

The continuous improvement of human-computer interaction technology makes it possible to compute emotions. In this paper, we introduce our submission to the CVPR 2023 Competition on Affective Behavior Analysis in-the-wild (ABAW). Sentiment analysis in human-computer interaction should, as far as possible Start with multiple dimensions, fill in the single imperfect emotion channel, and finally determine the emotion tendency by fitting multiple results. Therefore, We exploited multimodal features extracted from video of different lengths from the competition dataset, including audio, pose and images. Well-informed emotion representations drive us to propose a Attention-based multimodal framework for emotion estimation. Our system achieves the performance of 0.361 on the validation dataset. The code is available at [https://github.com/xkwangcn/ABAW-5th-RT-IAI].

cs.CV↗

arXiv2023

Extreme Masking for Learning Instance and Distributed Visual Representations

The paper presents a scalable approach for learning spatially distributed visual representations over individual tokens and a holistic instance representation simultaneously. We use self-attention blocks to represent spatially distributed tokens, followed by cross-attention blocks to aggregate the holistic image instance. The core of the approach is the use of extremely large token masking (75\%-90\%) as the data augmentation for supervision. Our model, named ExtreMA, follows the plain BYOL approach where the instance representation from the unmasked subset is trained to predict that from the intact input. Instead of encouraging invariance across inputs, the model is required to capture informative variations in an image. The paper makes three contributions: 1) It presents random masking as a strong and computationally efficient data augmentation for siamese representation learning. 2) With multiple sampling per instance, extreme masking greatly speeds up learning and improves performance with more data. 3) ExtreMA obtains stronger linear probing performance than masked modeling methods, and better transfer performance than prior contrastive models.

cs.CV↗

arXiv2022

Regrowth-free AlGaInAs MQW polarization controller integrated with sidewall grating DFB laser

We report an AlGaInAs multiple quantum well integrated source of polarization controlled light consisting of a polarization mode converter PMC, differential phase shifter(DPS), and a side wall grating distributed-feedback DFB laser. We demonstrate an asymmetrical stepped-height ridge waveguide PMC to realize TE to TM polarization conversion and a symmetrical straight waveguide DPS to enable polarization rotation from approximately counterclockwise circular polarization to linear polarization. Based on the identical epitaxial layer scheme, all of the PMC, DPS, and DFB laser can be integrated monolithically using only a single step of metalorganic vapor phase epitaxy and two steps of III V material dry etching. For the DFB-PMC device, a high TE to TM polarization conversion efficiency 98% over a wide range of DFB injection currents is reported at 1555 nm wavelength. For the DFB-PMC-DPS device, a 60 degree rotation of the Stokes vector was obtained on the Poincaré sphere with a range of bias voltage from 0 V to -4.0 V at IDFB is 170 mA.

physics.optics↗

arXiv2022

Power Efficient Video Super-Resolution on Mobile NPUs with Deep Learning, Mobile AI & AIM 2022 challenge: Report

Video super-resolution is one of the most popular tasks on mobile devices, being widely used for an automatic improvement of low-bitrate and low-resolution video streams. While numerous solutions have been proposed for this problem, they are usually quite computationally demanding, demonstrating low FPS rates and power efficiency on mobile devices. In this Mobile AI challenge, we address this problem and propose the participants to design an end-to-end real-time video super-resolution solution for mobile NPUs optimized for low energy consumption. The participants were provided with the REDS training dataset containing video sequences for a 4X video upscaling task. The runtime and power efficiency of all models was evaluated on the powerful MediaTek Dimensity 9000 platform with a dedicated AI processing unit capable of accelerating floating-point and quantized neural networks. All proposed solutions are fully compatible with the above NPU, demonstrating an up to 500 FPS rate and 0.2 [Watt / 30 FPS] power consumption. A detailed description of all models developed in the challenge is provided in this paper.

eess.IV↗

arXiv2022

Stepped-height ridge waveguide MQW polarization mode converter monolithically integrated with sidewall grating DFB laser

We report the first demonstration of a 1555 nm stepped-height ridge waveguide polarization mode converter monolithically integrated with a side wall grating distributed-feedback (DFB) laser using the identical epitaxial layer scheme. The device shows stable single longitudinal mode (SLM) operation with the output light converted from TE to TM polarization with an efficiency of >94% over a wide range of DFB injection currents (IDFB) from 140 mA to 190 mA. The highest TM mode purity of 98.2% was obtained at IDFB=180 mA. A particular advantage of this device is that only a single step of metalorganic vapor-phase epitaxy and two steps of III-V material dry etching are required for the whole integrated device fabrication, significantly reducing complexity and cost.

physics.optics↗

arXiv2022

Unsupervised Learning of Efficient Geometry-Aware Neural Articulated Representations

We propose an unsupervised method for 3D geometry-aware representation learning of articulated objects, in which no image-pose pairs or foreground masks are used for training. Though photorealistic images of articulated objects can be rendered with explicit pose control through existing 3D neural representations, these methods require ground truth 3D pose and foreground masks for training, which are expensive to obtain. We obviate this need by learning the representations with GAN training. The generator is trained to produce realistic images of articulated objects from random poses and latent vectors by adversarial training. To avoid a high computational cost for GAN training, we propose an efficient neural representation for articulated objects based on tri-planes and then present a GAN-based framework for its unsupervised training. Experiments demonstrate the efficiency of our method and show that GAN-based training enables the learning of controllable 3D representations without paired supervision.

cs.CV↗

arXiv2022

Hybrid Multimodal Fusion for Humor Detection

In this paper, we present our solution to the MuSe-Humor sub-challenge of the Multimodal Emotional Challenge (MuSe) 2022. The goal of the MuSe-Humor sub-challenge is to detect humor and calculate AUC from audiovisual recordings of German football Bundesliga press conferences. It is annotated for humor displayed by the coaches. For this sub-challenge, we first build a discriminant model using the transformer module and BiLSTM module, and then propose a hybrid fusion strategy to use the prediction results of each modality to improve the performance of the model. Our experiments demonstrate the effectiveness of our proposed model and hybrid fusion strategy on multimodal fusion, and the AUC of our proposed model on the test set is 0.8972.

cs.LG↗

arXiv2022

Hybrid Multimodal Feature Extraction, Mining and Fusion for Sentiment Analysis

In this paper, we present our solutions for the Multimodal Sentiment Analysis Challenge (MuSe) 2022, which includes MuSe-Humor, MuSe-Reaction and MuSe-Stress Sub-challenges. The MuSe 2022 focuses on humor detection, emotional reactions and multimodal emotional stress utilizing different modalities and data sets. In our work, different kinds of multimodal features are extracted, including acoustic, visual, text and biological features. These features are fused by TEMMA and GRU with self-attention mechanism frameworks. In this paper, 1) several new audio features, facial expression features and paragraph-level text embeddings are extracted for accuracy improvement. 2) we substantially improve the accuracy and reliability of multimodal sentiment prediction by mining and blending the multimodal features. 3) effective data augmentation strategies are applied in model training to alleviate the problem of sample imbalance and prevent the model from learning biased subject characters. For the MuSe-Humor sub-challenge, our model obtains the AUC score of 0.8932. For the MuSe-Reaction sub-challenge, the Pearson's Correlations Coefficient of our approach on the test set is 0.3879, which outperforms all other participants. For the MuSe-Stress sub-challenge, our approach outperforms the baseline in both arousal and valence on the test dataset, reaching a final combined result of 0.5151.

cs.CV↗

arXiv2022

Bringing Rolling Shutter Images Alive with Dual Reversed Distortion

Rolling shutter (RS) distortion can be interpreted as the result of picking a row of pixels from instant global shutter (GS) frames over time during the exposure of the RS camera. This means that the information of each instant GS frame is partially, yet sequentially, embedded into the row-dependent distortion. Inspired by this fact, we address the challenging task of reversing this process, i.e., extracting undistorted GS frames from images suffering from RS distortion. However, since RS distortion is coupled with other factors such as readout settings and the relative velocity of scene elements to the camera, models that only exploit the geometric correlation between temporally adjacent images suffer from poor generality in processing data with different readout settings and dynamic scenes with both camera motion and object motion. In this paper, instead of two consecutive frames, we propose to exploit a pair of images captured by dual RS cameras with reversed RS directions for this highly challenging task. Grounded on the symmetric and complementary nature of dual reversed distortion, we develop a novel end-to-end model, IFED, to generate dual optical flow sequence through iterative learning of the velocity field during the RS time. Extensive experimental results demonstrate that IFED is superior to naive cascade schemes, as well as the state-of-the-art which utilizes adjacent RS images. Most importantly, although it is trained on a synthetic dataset, IFED is shown to be effective at retrieving GS frame sequences from real-world RS distorted images of dynamic scenes. Code is available at https://github.com/zzh-tech/Dual-Reversed-RS.

cs.CV↗

arXiv2022

Animation from Blur: Multi-modal Blur Decomposition with Motion Guidance

We study the challenging problem of recovering detailed motion from a single motion-blurred image. Existing solutions to this problem estimate a single image sequence without considering the motion ambiguity for each region. Therefore, the results tend to converge to the mean of the multi-modal possibilities. In this paper, we explicitly account for such motion ambiguity, allowing us to generate multiple plausible solutions all in sharp detail. The key idea is to introduce a motion guidance representation, which is a compact quantization of 2D optical flow with only four discrete motion directions. Conditioned on the motion guidance, the blur decomposition is led to a specific, unambiguous solution by using a novel two-stage decomposition network. We propose a unified framework for blur decomposition, which supports various interfaces for generating our motion guidance, including human input, motion information from adjacent video frames, and learning from a video dataset. Extensive experiments on synthesized datasets and real-world data show that the proposed framework is qualitatively and quantitatively superior to previous methods, and also offers the merit of producing physically plausible and diverse solutions. Code is available at https://github.com/zzh-tech/Animation-from-Blur.

cs.CV↗

arXiv2022

Testing gravitational redshift based on microwave frequency links onboard China Space Station

In 2022 China Space Station (CSS) will be equipped with atomic clocks and optical clocks with stabilities of $2 \times 10^{-16}$ and $8 \times 10^{-18}$, respectively, which provides an excellent opportunity to test gravitational redshift (GR) with higher accuracy than previous results. Based on high-precise frequency links between CSS and a ground station, we formulated a model and provided simulation experiments to test GR. Simulation results suggest that this method could test the GR at the accuracy level of $(0.27 \pm 2.15) \times10^{-7}$, more than two orders in magnitude higher than the result of the experiment of a hydrogen clock on board a flying rocket more than 40 years ago.

astro-ph.IM↗

arXiv2022

Cross-Model Pseudo-Labeling for Semi-Supervised Action Recognition

Semi-supervised action recognition is a challenging but important task due to the high cost of data annotation. A common approach to this problem is to assign unlabeled data with pseudo-labels, which are then used as additional supervision in training. Typically in recent work, the pseudo-labels are obtained by training a model on the labeled data, and then using confident predictions from the model to teach itself. In this work, we propose a more effective pseudo-labeling scheme, called Cross-Model Pseudo-Labeling (CMPL). Concretely, we introduce a lightweight auxiliary network in addition to the primary backbone, and ask them to predict pseudo-labels for each other. We observe that, due to their different structural biases, these two models tend to learn complementary representations from the same video clips. Each model can thus benefit from its counterpart by utilizing cross-model predictions as supervision. Experiments on different data partition protocols demonstrate the significant improvement of our framework over existing alternatives. For example, CMPL achieves $17.6\%$ and $25.1\%$ Top-1 accuracy on Kinetics-400 and UCF-101 using only the RGB modality and $1\%$ labeled data, outperforming our baseline model, FixMatch, by $9.0\%$ and $10.3\%$, respectively.

cs.CV↗

arXiv2022

Determining the gravity potential with the CVSTT technique using two hydrogen clocks

According to general relativity theory (GRT), by comparing the frequencies between two precise clocks at two different stations, the gravity potential (geopotential) difference between the two stations can be determined due to the gravity frequency shift effect. Here, we provide experimental results of geopotential difference determination based on frequency comparisons between two remote hydrogen atomic clocks, with the help of common-view satellite time transfer (CVSTT) technique. For the first time we apply the ensemble empirical mode decomposition (EEMD) technique to the CVSTT observations for effectively determining the geopotential-related signals. Based on the net frequency shift between the two clocks in two different periods, the geopotential difference between stations of the Beijing 203 Institute Laboratory (BIL) and Luojiashan Time--Frequency Station (LTS) is determined. Comparisons show that the orthometric height (OH) of LTS determined by the clock comparison is deviated from that determined by the Earth gravity model EGM2008 by (38.5$\pm$45.7)~m. The results are consistent with the frequency stabilities of the hydrogen clocks (at the level of $10^{-15}$~day$^{-1}$) used in the experiment. Using more precise atomic or optical clocks, the CVSTT method for geopotential determination could be applied effectively and extensively in geodesy in the future.

physics.geo-ph↗