Multimodal Language Learning Models (MLLMs) have shown remarkable performance in image understanding, generation, and editing, with recent advancements achieving pixel-level grounding with reasoning. However, these models for common objects struggle with fine-grained face understanding. In this work, we introduce the FacePlayGround-240K dataset, the first pioneering large-scale, pixel-grounded face caption and question-answer (QA) dataset that includes 240K images, 47 mask categories, 5.4M mask annotations, and 7.3M grounded regions, meticulously curated for alignment pretraining and instruction-tuning. We present the GroundingFace framework, specifically designed to enhance fine-grained face understanding. This framework significantly augments the capabilities of existing grounding models in face part segmentation, face attribute comprehension, while preserving general scene understanding. Comprehensive experiments validate that our approach surpasses current state-of-the-art models in pixel-grounded face captioning/QA and various downstream tasks, including face captioning, referring segmentation, and zero-shot face attribute recognition.

The study of talking face generation mainly explores the intricacies of synchronizing facial movements and crafting visually appealing, temporally-coherent animations. However, due to the limited exploration of global audio perception, current approaches predominantly employ auxiliary visual and spatial knowledge to stabilize the movements, which often results in the deterioration of the naturalness and temporal inconsistencies. Considering the essence of audio-driven animation, the audio signal serves as the ideal and unique priors to adjust facial expressions and lip movements, without resorting to interference of any visual signals. Based on this motivation, we propose a novel paradigm, dubbed as Sonic, to shift focus on the exploration of global audio perception. To effectively leverage global audio knowledge, we disentangle it into intra-and inter-clip audio perception and collaborate with both aspects to enhance overall perception. For the intra-clip audio perception, 1). Context-enhanced audio learning, in which long-range intra-clip temporal audio knowledge is extracted to provide facial expression and lip motion priors implicitly expressed as the tone and speed of speech. 2). Motion-decoupled controller, in which the motion of the head and expression movement are disentangled and independently controlled by intra-audio clips. Most importantly, for inter-clip audio perception, as a bridge to connect the intra-clips to achieve the global perception, Time-aware position shift fusion, in which the global inter-clip audio information is considered and fused for long-audio inference via through consecutively time-aware shifted windows. Extensive experiments demonstrate that the novel audio-driven paradigm outperform existing SOTA methodologies in terms of video quality, temporally consistency, lip synchronization precision, and motion diversity.

Recent advances in Talking Head Generation (THG) have achieved impressive lip synchronization and visual quality through diffusion models; yet existing methods struggle to generate emotionally expressive portraits while preserving speaker identity. We identify three critical limitations in current emotional talking head generation: insufficient utilization of audio's inherent emotional cues, identity leakage in emotion representations, and isolated learning of emotion correlations. To address these challenges, we propose a novel framework dubbed as DICE-Talk, following the idea of disentangling identity with emotion, and then cooperating emotions with similar characteristics. First, we develop a disentangled emotion embedder that jointly models audio-visual emotional cues through cross-modal attention, representing emotions as identity-agnostic Gaussian distributions. Second, we introduce a correlation-enhanced emotion conditioning module with learnable emotion banks that explicitly capture inter-emotion relationships through vector quantization and attention-based feature aggregation. Third, we design an emotion discrimination objective that enforces affective consistency during the diffusion process through latent-space classification. Extensive experiments on MEAD and HDTF datasets demonstrate our method's superiority, outperforming state-of-the-art approaches in emotion accuracy while maintaining competitive lip-sync performance. Qualitative results and user studies further confirm our method's ability to generate identity-preserving portraits with rich, correlated emotional expressions that naturally adapt to unseen identities.

The last decades are marked by massive and diverse image data, which shows increasingly high resolution and quality. However, some images we obtained may be corrupted, affecting the perception and the application of downstream tasks. A generic method for generating a high-quality image from the degraded one is in demand. In this paper, we present a novel GAN inversion framework that utilizes the powerful generative ability of StyleGAN-XL for this problem. To ease the inversion challenge with StyleGAN-XL, Clustering \& Regularize Inversion (CRI) is proposed. Specifically, the latent space is firstly divided into finer-grained sub-spaces by clustering. Instead of initializing the inversion with the average latent vector, we approximate a centroid latent vector from the clusters, which generates an image close to the input image. Then, an offset with a regularization term is introduced to keep the inverted latent vector within a certain range. We validate our CRI scheme on multiple restoration tasks (i.e., inpainting, colorization, and super-resolution) of complex natural images, and show preferable quantitative and qualitative results. We further demonstrate our technique is robust in terms of data and different GAN models. To our best knowledge, we are the first to adopt StyleGAN-XL for generating high-quality natural images from diverse degraded inputs. Code is available at https://github.com/Booooooooooo/CRI.

To achieve promising results on blind image super-resolution (SR),some attempts leveraged the low resolution (LR) images to predict the kernel and improve the SR performance. However, these Supervised Kernel Prediction (SKP) methods are impractical due to the unavailable real-world blur kernels. Although some Unsupervised Degradation Prediction (UDP) methods are proposed to bypass this problem, the inconsistency between degradation embedding and SR feature is still challenging. By exploring the correlations between degradation embedding and SR feature, we observe that jointly learning the content and degradation aware feature is optimal. Based on this observation, a Content and Degradation aware SR Network dubbed CDSR is proposed. Specifically, CDSR contains three newly-established modules: (1) a Lightweight Patch-based Encoder (LPE) is applied to jointly extract content and degradation features; (2) a Domain Query Attention based module (DQA) is employed to adaptively reduce the inconsistency; (3) a Codebook-based Space Compress module (CSC) that can suppress the redundant information. Extensive experiments on several benchmarks demonstrate that the proposed CDSR outperforms the existing UDP models and achieves competitive performance on PSNR and SSIM even compared with the state-of-the-art SKP methods.

Temporal action localization is an important yet challenging task in video understanding. Typically, such a task aims at inferring both the action category and localization of the start and end frame for each action instance in a long, untrimmed video. While most current models achieve good results by using pre-defined anchors and numerous actionness, such methods could be bothered with both large number of outputs and heavy tuning of locations and sizes corresponding to different anchors. Instead, anchor-free methods is lighter, getting rid of redundant hyper-parameters, but gains few attention. In this paper, we propose the first purely anchor-free temporal localization method, which is both efficient and effective. Our model includes (i) an end-to-end trainable basic predictor, (ii) a saliency-based refinement module to gather more valuable boundary features for each proposal with a novel boundary pooling, and (iii) several consistency constraints to make sure our model can find the accurate boundary given arbitrary proposals. Extensive experiments show that our method beats all anchor-based and actionness-guided methods with a remarkable margin on THUMOS14, achieving state-of-the-art results, and comparable ones on ActivityNet v1.3. Code is available at https://github.com/TencentYoutuResearch/ActionDetection-AFSD.

Deep-learning based Super-Resolution (SR) methods have exhibited promising performance under non-blind setting where blur kernel is known; however, blur kernels of Low-Resolution (LR) images in different practical applications are usually unknown. It may lead to a significant performance drop when degradation process of training images deviates from that of real images. In this paper, we propose a novel blind SR framework to super-resolve LR images degraded by arbitrary blur kernel with accurate kernel estimation in frequency domain. To our best knowledge, this is the first deep learning method which conducts blur kernel estimation in frequency domain. Specifically, we first demonstrate that feature representation in frequency domain is more conducive for blur kernel reconstruction than in spatial domain. Next, we present a Spectrum-to-Kernel (S$2$K) network to estimate general blur kernels in diverse forms. We use a conditional GAN (CGAN) combined with SR-oriented optimization target to learn the end-to-end translation from degraded images' spectra to unknown kernels. Extensive experiments on both synthetic and real-world images demonstrate that our proposed method sufficiently reduces blur kernel estimation error, thus enables the off-the-shelf non-blind SR methods to work under blind setting effectively, and achieves superior performance over state-of-the-art blind SR methods, averagely by 1.39dB, 0.48dB (Gaussian kernels) and 6.15dB, 4.57dB (motion kernels) for scales $2\times$ and $4\times$ respectively.

Generating temporal action proposals remains a very challenging problem, where the main issue lies in predicting precise temporal proposal boundaries and reliable action confidence in long and untrimmed real-world videos. In this paper, we propose an efficient and unified framework to generate temporal action proposals named Dense Boundary Generator (DBG), which draws inspiration from boundary-sensitive methods and implements boundary classification and action completeness regression for densely distributed proposals. In particular, the DBG consists of two modules: Temporal boundary classification (TBC) and Action-aware completeness regression (ACR). The TBC aims to provide two temporal boundary confidence maps by low-level two-stream features, while the ACR is designed to generate an action completeness score map by high-level action-aware features. Moreover, we introduce a dual stream BaseNet (DSB) to encode RGB and optical flow information, which helps to capture discriminative boundary and actionness features. Extensive experiments on popular benchmarks ActivityNet-1.3 and THUMOS14 demonstrate the superiority of DBG over the state-of-the-art proposal generator (e.g., MGG and BMN).

Supervised deep learning techniques have achieved great success in various fields due to getting rid of the limitation of handcrafted representations. However, most previous image retargeting algorithms still employ fixed design principles such as using gradient map or handcrafted features to compute saliency map, which inevitably restricts its generality. Deep learning techniques may help to address this issue, but the challenging problem is that we need to build a large-scale image retargeting dataset for the training of deep retargeting models. However, building such a dataset requires huge human efforts. In this paper, we propose a novel deep cyclic image retargeting approach, called Cycle-IR, to firstly implement image retargeting with a single deep model, without relying on any explicit user annotations. Our idea is built on the reverse mapping from the retargeted images to the given images. If the retargeted image has serious distortion or excessive loss of important visual information, the reverse mapping is unlikely to restore the input image well. We constrain this forward-reverse consistency by introducing a cyclic perception coherence loss. In addition, we propose a simple yet effective image retargeting network (IRNet) to implement the image retargeting process. Our IRNet contains a spatial and channel attention layer, which is able to discriminate visually important regions of input images effectively, especially in cluttered images. Given arbitrary sizes of input images and desired aspect ratios, our Cycle-IR can produce visually pleasing target images directly. Extensive experiments on the standard RetargetMe dataset show the superiority of our Cycle-IR.

For video super-resolution, current state-of-the-art approaches either process multiple low-resolution (LR) frames to produce each output high-resolution (HR) frame separately in a sliding window fashion or recurrently exploit the previously estimated HR frames to super-resolve the following frame. The main weaknesses of these approaches are: 1) separately generating each output frame may obtain high-quality HR estimates while resulting in unsatisfactory flickering artifacts, and 2) combining previously generated HR frames can produce temporally consistent results in the case of short information flow, but it will cause significant jitter and jagged artifacts because the previous super-resolving errors are constantly accumulated to the subsequent frames.In this paper, we propose a fully end-to-end trainable frame and feature-context video super-resolution (FFCVSR) network that consists of two key sub-networks: local network and context network, where the first one explicitly utilizes a sequence of consecutive LR frames to generate local feature and local SR frame, and the other combines the outputs of local network and the previously estimated HR frames and features to super-resolve the subsequent frame. Our approach takes full advantage of the inter-frame information from multiple LR frames and the context information from previously predicted HR frames, producing temporally consistent highquality results while maintaining real-time speed by directly reusing previous features and frames. Extensive evaluations and comparisons demonstrate that our approach produces state-of-the-art results on a standard benchmark dataset, with advantages in terms of accuracy, efficiency, and visual quality over the existing approaches.

Foreground detection is an important part of surveillance video analysis, and also has challenges. For example, the classical methods are difficult to distinguish the foreground, which is similar to the background. In recent years, Convolutional Neural Networks (CNNs) have been widely used in image processing and achieved better performance. In this paper, we proposed an efficient deep Fully Convolutional Semantic Networks (FCSN) model for foreground detection in surveillance video. Our model aimed at learning the global differences between the video frame and the background image, and the semantic information by utilizing the pre-trained weights on semantic segmentation. In the experiment, unlike other related work, we proposed a reasonable method, which is able to avoid overfitting results to construct training data with 20 videos and test data with 6 videos on the dataset of 2014 ChangeDetection.net (CDnet 2014). Experimental results verified that our model outperforms the state-of-the-art methods in the foreground detection of surveillance video.

The popularity of mobile applications has greatly enriched and facilitated our lives. However, the rapid increase of digital images and the problem of narrow bandwidth of the wireless network call for an appropriate approach to reduce the amount of data transmitted over the wireless network (i.e., low bit-rate transmission) while ensuring high recognition accuracy at the cloud. We propose a simple and effective feature retargeting (FR) approach for retargeting an image while preserving the representative local features (e.g., SIFT, SURF, and BRIEF) in the image. Our feature retargeting approach aims at low bit-rate visual recognition instead of high-quality visual perception that visual retargeting methods dedicate to. Our algorithm consists of two key novelties: estimating feature saliency and retargeting image: Estimating feature saliency focuses on predicting the relative importance of different features in an image by analyzing uniqueness in a specific context; Retargeting image aims at finding the optimal resolution for the retargeted image to maximize feature-saliency energy. We evaluate the proposed approach for two different applications in three large data sets and observe that our FR approach consistently outperforms state-of-the-art retargeting algorithms, resulting in both higher precision and lower bit-rates. We also demonstrate that even when the resolution of source image is reduced greatly, e.g., 1/7 original size, our algorithm produces superior results as compared with other approaches.