Shanghua Gao

Shanghua Gao, Pan Zhou, Ming-Ming Cheng et al.

Despite its success in image synthesis, we observe that diffusion probabilistic models (DPMs) often lack contextual reasoning ability to learn the relations among object parts in an image, leading to a slow learning process. To solve this issue, we propose a Masked Diffusion Transformer (MDT) that introduces a mask latent modeling scheme to explicitly enhance the DPMs' ability to contextual relation learning among object semantic parts in an image. During training, MDT operates in the latent space to mask certain tokens. Then, an asymmetric diffusion transformer is designed to predict masked tokens from unmasked ones while maintaining the diffusion generation process. Our MDT can reconstruct the full information of an image from its incomplete contextual input, thus enabling it to learn the associated relations among image tokens. We further improve MDT with a more efficient macro network structure and training strategy, named MDTv2. Experimental results show that MDTv2 achieves superior image synthesis performance, e.g., a new SOTA FID score of 1.58 on the ImageNet dataset, and has more than 10x faster learning speed than the previous SOTA DiT. The source code is released at https://github.com/sail-sg/MDT.

Zhong-Yu Li, Shanghua Gao, Ming-Ming Cheng

Learning representations with self-supervision for convolutional networks (CNN) has been validated to be effective for vision tasks. As an alternative to CNN, vision transformers (ViT) have strong representation ability with spatial self-attention and channel-level feedforward networks. Recent works reveal that self-supervised learning helps unleash the great potential of ViT. Still, most works follow self-supervised strategies designed for CNN, e.g., instance-level discrimination of samples, but they ignore the properties of ViT. We observe that relational modeling on spatial and channel dimensions distinguishes ViT from other networks. To enforce this property, we explore the feature SElf-RElation (SERE) for training self-supervised ViT. Specifically, instead of conducting self-supervised learning solely on feature embeddings from multiple views, we utilize the feature self-relations, i.e., spatial/channel self-relations, for self-supervised learning. Self-relation based learning further enhances the relation modeling ability of ViT, resulting in stronger representations that stably improve performance on multiple downstream tasks. Our source code is publicly available at: https://github.com/MCG-NKU/SERE.

Shanghua Gao, Zhong-Yu Li, Qi Han et al.

Temporal/spatial receptive fields of models play an important role in sequential/spatial tasks. Large receptive fields facilitate long-term relations, while small receptive fields help to capture the local details. Existing methods construct models with hand-designed receptive fields in layers. Can we effectively search for receptive field combinations to replace hand-designed patterns? To answer this question, we propose to find better receptive field combinations through a global-to-local search scheme. Our search scheme exploits both global search to find the coarse combinations and local search to get the refined receptive field combinations further. The global search finds possible coarse combinations other than human-designed patterns. On top of the global search, we propose an expectation-guided iterative local search scheme to refine combinations effectively. Our RF-Next models, plugging receptive field search to various models, boost the performance on many tasks, e.g., temporal action segmentation, object detection, instance segmentation, and speech synthesis. The source code is publicly available on http://mmcheng.net/rfnext.

Yilin Zhao, Xinbin Yuan, Shanghua Gao et al.

In this technical report, we target generating anthropomorphized personas for LLM-based characters in an online manner, including visual appearance, personality and tones, with only text descriptions. To achieve this, we first leverage the in-context learning capability of LLMs for personality generation by carefully designing a set of system prompts. We then propose two novel concepts: the mixture of voices (MoV) and the mixture of diffusers (MoD) for diverse voice and appearance generation. For MoV, we utilize the text-to-speech (TTS) algorithms with a variety of pre-defined tones and select the most matching one based on the user-provided text description automatically. For MoD, we combine the recent popular text-to-image generation techniques and talking head algorithms to streamline the process of generating talking objects. We termed the whole framework as ChatAnything. With it, users could be able to animate anything with any personas that are anthropomorphic using just a few text inputs. However, we have observed that the anthropomorphic objects produced by current generative models are often undetectable by pre-trained face landmark detectors, leading to failure of the face motion generation, even if these faces possess human-like appearances because those images are nearly seen during the training (e.g., OOD samples). To address this issue, we incorporate pixel-level guidance to infuse human face landmarks during the image generation phase. To benchmark these metrics, we have built an evaluation dataset. Based on it, we verify that the detection rate of the face landmark is significantly increased from 57.0% to 92.5% thus allowing automatic face animation based on generated speech content. The code and more results can be found at https://chatanything.github.io/.

Shanghua Gao, Pan Zhou, Ming-Ming Cheng et al.

Although increasingly training-expensive, most self-supervised learning (SSL) models have repeatedly been trained from scratch but not fully utilized, since only a few SOTAs are employed for downstream tasks. In this work, we explore a sustainable SSL framework with two major challenges: i) learning a stronger new SSL model based on the existing pretrained SSL model, also called as "base" model, in a cost-friendly manner, ii) allowing the training of the new model to be compatible with various base models. We propose a Target-Enhanced Conditional (TEC) scheme which introduces two components to the existing mask-reconstruction based SSL. Firstly, we propose patch-relation enhanced targets which enhances the target given by base model and encourages the new model to learn semantic-relation knowledge from the base model by using incomplete inputs. This hardening and target-enhancing help the new model surpass the base model, since they enforce additional patch relation modeling to handle incomplete input. Secondly, we introduce a conditional adapter that adaptively adjusts new model prediction to align with the target of different base models. Extensive experimental results show that our TEC scheme can accelerate the learning speed, and also improve SOTA SSL base models, e.g., MAE and iBOT, taking an explorative step towards sustainable SSL.

Ge Wu, Shen Zhang, Ruijing Shi et al.

REPA and its variants effectively mitigate training challenges in diffusion models by incorporating external visual representations from pretrained models, through alignment between the noisy hidden projections of denoising networks and foundational clean image representations. We argue that the external alignment, which is absent during the entire denoising inference process, falls short of fully harnessing the potential of discriminative representations. In this work, we propose a straightforward method called Representation Entanglement for Generation (REG), which entangles low-level image latents with a single high-level class token from pretrained foundation models for denoising. REG acquires the capability to produce coherent image-class pairs directly from pure noise, substantially improving both generation quality and training efficiency. This is accomplished with negligible additional inference overhead, requiring only one single additional token for denoising (<0.5\% increase in FLOPs and latency). The inference process concurrently reconstructs both image latents and their corresponding global semantics, where the acquired semantic knowledge actively guides and enhances the image generation process. On ImageNet 256$\times$256, SiT-XL/2 + REG demonstrates remarkable convergence acceleration, achieving $\textbf{63}\times$ and $\textbf{23}\times$ faster training than SiT-XL/2 and SiT-XL/2 + REPA, respectively. More impressively, SiT-L/2 + REG trained for merely 400K iterations outperforms SiT-XL/2 + REPA trained for 4M iterations ($\textbf{10}\times$ longer). Code is available at: https://github.com/Martinser/REG.

Yunheng Li, Zhongyu Li, Shanghua Gao et al.

Effectively modeling discriminative spatio-temporal information is essential for segmenting activities in long action sequences. However, we observe that existing methods are limited in weak spatio-temporal modeling capability due to two forms of decoupled modeling: (i) cascaded interaction couples spatial and temporal modeling, which over-smooths motion modeling over the long sequence, and (ii) joint-shared temporal modeling adopts shared weights to model each joint, ignoring the distinct motion patterns of different joints. We propose a Decoupled Spatio-Temporal Framework (DeST) to address the above issues. Firstly, we decouple the cascaded spatio-temporal interaction to avoid stacking multiple spatio-temporal blocks, while achieving sufficient spatio-temporal interaction. Specifically, DeST performs once unified spatial modeling and divides the spatial features into different groups of subfeatures, which then adaptively interact with temporal features from different layers. Since the different sub-features contain distinct spatial semantics, the model could learn the optimal interaction pattern at each layer. Meanwhile, inspired by the fact that different joints move at different speeds, we propose joint-decoupled temporal modeling, which employs independent trainable weights to capture distinctive temporal features of each joint. On four large-scale benchmarks of different scenes, DeST significantly outperforms current state-of-the-art methods with less computational complexity.

Duo Li, Shang-Hua Gao

In recent years, the connections between deep residual networks and first-order Ordinary Differential Equations (ODEs) have been disclosed. In this work, we further bridge the deep neural architecture design with the second-order ODEs and propose a novel reversible neural network, termed as m-RevNet, that is characterized by inserting momentum update to residual blocks. The reversible property allows us to perform backward pass without access to activation values of the forward pass, greatly relieving the storage burden during training. Furthermore, the theoretical foundation based on second-order ODEs grants m-RevNet with stronger representational power than vanilla residual networks, which potentially explains its performance gains. For certain learning scenarios, we analytically and empirically reveal that our m-RevNet succeeds while standard ResNet fails. Comprehensive experiments on various image classification and semantic segmentation benchmarks demonstrate the superiority of our m-RevNet over ResNet, concerning both memory efficiency and recognition performance.

Shanghua Gao, Zhong-Yu Li, Ming-Hsuan Yang et al.

Empowered by large datasets, e.g., ImageNet, unsupervised learning on large-scale data has enabled significant advances for classification tasks. However, whether the large-scale unsupervised semantic segmentation can be achieved remains unknown. There are two major challenges: i) we need a large-scale benchmark for assessing algorithms; ii) we need to develop methods to simultaneously learn category and shape representation in an unsupervised manner. In this work, we propose a new problem of large-scale unsupervised semantic segmentation (LUSS) with a newly created benchmark dataset to help the research progress. Building on the ImageNet dataset, we propose the ImageNet-S dataset with 1.2 million training images and 50k high-quality semantic segmentation annotations for evaluation. Our benchmark has a high data diversity and a clear task objective. We also present a simple yet effective method that works surprisingly well for LUSS. In addition, we benchmark related un/weakly/fully supervised methods accordingly, identifying the challenges and possible directions of LUSS. The benchmark and source code is publicly available at https://github.com/LUSSeg.

Shang-Hua Gao, Qi Han, Zhong-Yu Li et al.

Temporal receptive fields of models play an important role in action segmentation. Large receptive fields facilitate the long-term relations among video clips while small receptive fields help capture the local details. Existing methods construct models with hand-designed receptive fields in layers. Can we effectively search for receptive field combinations to replace hand-designed patterns? To answer this question, we propose to find better receptive field combinations through a global-to-local search scheme. Our search scheme exploits both global search to find the coarse combinations and local search to get the refined receptive field combination patterns further. The global search finds possible coarse combinations other than human-designed patterns. On top of the global search, we propose an expectation guided iterative local search scheme to refine combinations effectively. Our global-to-local search can be plugged into existing action segmentation methods to achieve state-of-the-art performance.

Qiulin Zhang, Zhuqing Jiang, Qishuo Lu et al.

Many effective solutions have been proposed to reduce the redundancy of models for inference acceleration. Nevertheless, common approaches mostly focus on eliminating less important filters or constructing efficient operations, while ignoring the pattern redundancy in feature maps. We reveal that many feature maps within a layer share similar but not identical patterns. However, it is difficult to identify if features with similar patterns are redundant or contain essential details. Therefore, instead of directly removing uncertain redundant features, we propose a \textbf{sp}lit based \textbf{conv}olutional operation, namely SPConv, to tolerate features with similar patterns but require less computation. Specifically, we split input feature maps into the representative part and the uncertain redundant part, where intrinsic information is extracted from the representative part through relatively heavy computation while tiny hidden details in the uncertain redundant part are processed with some light-weight operation. To recalibrate and fuse these two groups of processed features, we propose a parameters-free feature fusion module. Moreover, our SPConv is formulated to replace the vanilla convolution in a plug-and-play way. Without any bells and whistles, experimental results on benchmarks demonstrate SPConv-equipped networks consistently outperform state-of-the-art baselines in both accuracy and inference time on GPU, with FLOPs and parameters dropped sharply.

Shang-Hua Gao, Yong-Qiang Tan, Ming-Ming Cheng et al.

Salient object detection models often demand a considerable amount of computation cost to make precise prediction for each pixel, making them hardly applicable on low-power devices. In this paper, we aim to relieve the contradiction between computation cost and model performance by improving the network efficiency to a higher degree. We propose a flexible convolutional module, namely generalized OctConv (gOctConv), to efficiently utilize both in-stage and cross-stages multi-scale features, while reducing the representation redundancy by a novel dynamic weight decay scheme. The effective dynamic weight decay scheme stably boosts the sparsity of parameters during training, supports learnable number of channels for each scale in gOctConv, allowing 80% of parameters reduce with negligible performance drop. Utilizing gOctConv, we build an extremely light-weighted model, namely CSNet, which achieves comparable performance with about 0.2% parameters (100k) of large models on popular salient object detection benchmarks.

Shang-Hua Gao, Ming-Ming Cheng, Kai Zhao et al.

Representing features at multiple scales is of great importance for numerous vision tasks. Recent advances in backbone convolutional neural networks (CNNs) continually demonstrate stronger multi-scale representation ability, leading to consistent performance gains on a wide range of applications. However, most existing methods represent the multi-scale features in a layer-wise manner. In this paper, we propose a novel building block for CNNs, namely Res2Net, by constructing hierarchical residual-like connections within one single residual block. The Res2Net represents multi-scale features at a granular level and increases the range of receptive fields for each network layer. The proposed Res2Net block can be plugged into the state-of-the-art backbone CNN models, e.g., ResNet, ResNeXt, and DLA. We evaluate the Res2Net block on all these models and demonstrate consistent performance gains over baseline models on widely-used datasets, e.g., CIFAR-100 and ImageNet. Further ablation studies and experimental results on representative computer vision tasks, i.e., object detection, class activation mapping, and salient object detection, further verify the superiority of the Res2Net over the state-of-the-art baseline methods. The source code and trained models are available on https://mmcheng.net/res2net/.

Kai Zhao, Shanghua Gao, Wenguan Wang et al.

Current CNN-based solutions to salient object detection (SOD) mainly rely on the optimization of cross-entropy loss (CELoss). Then the quality of detected saliency maps is often evaluated in terms of F-measure. In this paper, we investigate an interesting issue: can we consistently use the F-measure formulation in both training and evaluation for SOD? By reformulating the standard F-measure we propose the relaxed F-measure which is differentiable w.r.t the posterior and can be easily appended to the back of CNNs as the loss function. Compared to the conventional cross-entropy loss of which the gradients decrease dramatically in the saturated area, our loss function, named FLoss, holds considerable gradients even when the activation approaches the target. Consequently, the FLoss can continuously force the network to produce polarized activations. Comprehensive benchmarks on several popular datasets show that FLoss outperforms the state-of-the-art with a considerable margin. More specifically, due to the polarized predictions, our method is able to obtain high-quality saliency maps without carefully tuning the optimal threshold, showing significant advantages in real-world applications.

Deng-Ping Fan, Ming-Ming Cheng, Jiang-Jiang Liu et al.

We provide a comprehensive evaluation of salient object detection (SOD) models. Our analysis identifies a serious design bias of existing SOD datasets which assumes that each image contains at least one clearly outstanding salient object in low clutter. The design bias has led to a saturated high performance for state-of-the-art SOD models when evaluated on existing datasets. The models, however, still perform far from being satisfactory when applied to real-world daily scenes. Based on our analyses, we first identify 7 crucial aspects that a comprehensive and balanced dataset should fulfill. Then, we propose a new high quality dataset and update the previous saliency benchmark. Specifically, our SOC (Salient Objects in Clutter) dataset, includes images with salient and non-salient objects from daily object categories. Beyond object category annotations, each salient image is accompanied by attributes that reflect common challenges in real-world scenes. Finally, we report attribute-based performance assessment on our dataset.

Kai Zhao, Wei Shen, Shanghua Gao et al.

In natural images, the scales (thickness) of object skeletons may dramatically vary among objects and object parts, making object skeleton detection a challenging problem. We present a new convolutional neural network (CNN) architecture by introducing a novel hierarchical feature integration mechanism, named Hi-Fi, to address the skeleton detection problem. The proposed CNN-based approach has a powerful multi-scale feature integration ability that intrinsically captures high-level semantics from deeper layers as well as low-level details from shallower layers. % By hierarchically integrating different CNN feature levels with bidirectional guidance, our approach (1) enables mutual refinement across features of different levels, and (2) possesses the strong ability to capture both rich object context and high-resolution details. Experimental results show that our method significantly outperforms the state-of-the-art methods in terms of effectively fusing features from very different scales, as evidenced by a considerable performance improvement on several benchmarks.