Transformer has achieved impressive successes for various computer vision tasks. However, most of existing studies require to pretrain the Transformer backbone on a large-scale labeled dataset (e.g., ImageNet) for achieving satisfactory performance, which is usually unavailable for medical images. Additionally, due to the gap between medical and natural images, the improvement generated by the ImageNet pretrained weights significantly degrades while transferring the weights to medical image processing tasks. In this paper, we propose Bootstrap Own Latent of Transformer (BOLT), a self-supervised learning approach specifically for medical image classification with the Transformer backbone. Our BOLT consists of two networks, namely online and target branches, for self-supervised representation learning. Concretely, the online network is trained to predict the target network representation of the same patch embedding tokens with a different perturbation. To maximally excavate the impact of Transformer from limited medical data, we propose an auxiliary difficulty ranking task. The Transformer is enforced to identify which branch (i.e., online/target) is processing the more difficult perturbed tokens. Overall, the Transformer endeavours itself to distill the transformation-invariant features from the perturbed tokens to simultaneously achieve difficulty measurement and maintain the consistency of self-supervised representations. The proposed BOLT is evaluated on three medical image processing tasks, i.e., skin lesion classification, knee fatigue fracture grading and diabetic retinopathy grading. The experimental results validate the superiority of our BOLT for medical image classification, compared to ImageNet pretrained weights and state-of-the-art self-supervised learning approaches.

Model bias triggered by long-tailed data has been widely studied. However, measure based on the number of samples cannot explicate three phenomena simultaneously: (1) Given enough data, the classification performance gain is marginal with additional samples. (2) Classification performance decays precipitously as the number of training samples decreases when there is insufficient data. (3) Model trained on sample-balanced datasets still has different biases for different classes. In this work, we define and quantify the semantic scale of classes, which is used to measure the feature diversity of classes. It is exciting to find experimentally that there is a marginal effect of semantic scale, which perfectly describes the first two phenomena. Further, the quantitative measurement of semantic scale imbalance is proposed, which can accurately reflect model bias on multiple datasets, even on sample-balanced data, revealing a novel perspective for the study of class imbalance. Due to the prevalence of semantic scale imbalance, we propose semantic-scale-balanced learning, including a general loss improvement scheme and a dynamic re-weighting training framework that overcomes the challenge of calculating semantic scales in real-time during iterations. Comprehensive experiments show that dynamic semantic-scale-balanced learning consistently enables the model to perform superiorly on large-scale long-tailed and non-long-tailed natural and medical datasets, which is a good starting point for mitigating the prevalent but unnoticed model bias.

Patients take care of what their teeth will be like after the orthodontics. Orthodontists usually describe the expectation movement based on the original smile images, which is unconvincing. The growth of deep-learning generative models change this situation. It can visualize the outcome of orthodontic treatment and help patients foresee their future teeth and facial appearance. While previous studies mainly focus on 2D or 3D virtual treatment outcome (VTO) at a profile level, the problem of simulating treatment outcome at a frontal facial image is poorly explored. In this paper, we build an efficient and accurate system for simulating virtual teeth alignment effects in a frontal facial image. Our system takes a frontal face image of a patient with visible malpositioned teeth and the patient's 3D scanned teeth model as input, and progressively generates the visual results of the patient's teeth given the specific orthodontics planning steps from the doctor (i.e., the specification of translations and rotations of individual tooth). We design a multi-modal encoder-decoder based generative model to synthesize identity-preserving frontal facial images with aligned teeth. In addition, the original image color information is used to optimize the orthodontic outcomes, making the results more natural. We conduct extensive qualitative and clinical experiments and also a pilot study to validate our method.

Zero-Shot Learning has been a highlighted research topic in both vision and language areas. Recently, most existing methods adopt structured knowledge information to model explicit correlations among categories and use deep graph convolutional network to propagate information between different categories. However, it is difficult to add new categories to existing structured knowledge graph, and deep graph convolutional network suffers from over-smoothing problem. In this paper, we provide a new semantic enhanced knowledge graph that contains both expert knowledge and categories semantic correlation. Our semantic enhanced knowledge graph can further enhance the correlations among categories and make it easy to absorb new categories. To propagate information on the knowledge graph, we propose a novel Residual Graph Convolutional Network (ResGCN), which can effectively alleviate the problem of over-smoothing. Experiments conducted on the widely used large-scale ImageNet-21K dataset and AWA2 dataset show the effectiveness of our method, and establish a new state-of-the-art on zero-shot learning. Moreover, our results on the large-scale ImageNet-21K with various feature extraction networks show that our method has better generalization and robustness.

Reinforcement learning (RL) problems can be challenging without well-shaped rewards. Prior work on provably efficient RL methods generally proposes to address this issue with dedicated exploration strategies. However, another way to tackle this challenge is to reformulate it as a multi-task RL problem, where the task space contains not only the challenging task of interest but also easier tasks that implicitly function as a curriculum. Such a reformulation opens up the possibility of running existing multi-task RL methods as a more efficient alternative to solving a single challenging task from scratch. In this work, we provide a theoretical framework that reformulates a single-task RL problem as a multi-task RL problem defined by a curriculum. Under mild regularity conditions on the curriculum, we show that sequentially solving each task in the multi-task RL problem is more computationally efficient than solving the original single-task problem, without any explicit exploration bonuses or other exploration strategies. We also show that our theoretical insights can be translated into an effective practical learning algorithm that can accelerate curriculum learning on simulated robotic tasks.

In person re-identification (ReID) tasks, many works explore the learning of part features to improve the performance over global image features. Existing methods extract part features in an explicit manner, by either using a hand-designed image division or keypoints obtained with external visual systems. In this work, we propose to learn Discriminative implicit Parts (DiPs) which are decoupled from explicit body parts. Therefore, DiPs can learn to extract any discriminative features that can benefit in distinguishing identities, which is beyond predefined body parts (such as accessories). Moreover, we propose a novel implicit position to give a geometric interpretation for each DiP. The implicit position can also serve as a learning signal to encourage DiPs to be more position-equivariant with the identity in the image. Lastly, a set of attributes and auxiliary losses are introduced to further improve the learning of DiPs. Extensive experiments show that the proposed method achieves state-of-the-art performance on multiple person ReID benchmarks.

Blind watermarking provides powerful evidence for copyright protection, image authentication, and tampering identification. However, it remains a challenge to design a watermarking model with high imperceptibility and robustness against strong noise attacks. To resolve this issue, we present a framework Combining the Invertible and Non-invertible (CIN) mechanisms. The CIN is composed of the invertible part to achieve high imperceptibility and the non-invertible part to strengthen the robustness against strong noise attacks. For the invertible part, we develop a diffusion and extraction module (DEM) and a fusion and split module (FSM) to embed and extract watermarks symmetrically in an invertible way. For the non-invertible part, we introduce a non-invertible attention-based module (NIAM) and the noise-specific selection module (NSM) to solve the asymmetric extraction under a strong noise attack. Extensive experiments demonstrate that our framework outperforms the current state-of-the-art methods of imperceptibility and robustness significantly. Our framework can achieve an average of 99.99% accuracy and 67.66 dB PSNR under noise-free conditions, while 96.64% and 39.28 dB combined strong noise attacks. The code will be available in https://github.com/rmpku/CIN.

Vision Transformers have shown great promise recently for many vision tasks due to the insightful architecture design and attention mechanism. By revisiting the self-attention responses in Transformers, we empirically observe two interesting issues. First, Vision Transformers present a queryirrelevant behavior at deep layers, where the attention maps exhibit nearly consistent contexts in global scope, regardless of the query patch position (also head-irrelevant). Second, the attention maps are intrinsically sparse, few tokens dominate the attention weights; introducing the knowledge from ConvNets would largely smooth the attention and enhance the performance. Motivated by above observations, we generalize self-attention formulation to abstract a queryirrelevant global context directly and further integrate the global context into convolutions. The resulting model, a Fully Convolutional Vision Transformer (i.e., FCViT), purely consists of convolutional layers and firmly inherits the merits of both attention mechanism and convolutions, including dynamic property, weight sharing, and short- and long-range feature modeling, etc. Experimental results demonstrate the effectiveness of FCViT. With less than 14M parameters, our FCViT-S12 outperforms related work ResT-Lite by 3.7% top1 accuracy on ImageNet-1K. When scaling FCViT to larger models, we still perform better than previous state-of-the-art ConvNeXt with even fewer parameters. FCViT-based models also demonstrate promising transferability to downstream tasks, like object detection, instance segmentation, and semantic segmentation. Codes and models are made available at: https://github.com/ma-xu/FCViT.

Conversational recommender systems (CRSs) often utilize external knowledge graphs (KGs) to introduce rich semantic information and recommend relevant items through natural language dialogues. However, original KGs employed in existing CRSs are often incomplete and sparse, which limits the reasoning capability in recommendation. Moreover, only few of existing studies exploit the dialogue context to dynamically refine knowledge from KGs for better recommendation. To address the above issues, we propose the Variational Reasoning over Incomplete KGs Conversational Recommender (VRICR). Our key idea is to incorporate the large dialogue corpus naturally accompanied with CRSs to enhance the incomplete KGs; and perform dynamic knowledge reasoning conditioned on the dialogue context. Specifically, we denote the dialogue-specific subgraphs of KGs as latent variables with categorical priors for adaptive knowledge graphs refactor. We propose a variational Bayesian method to approximate posterior distributions over dialogue-specific subgraphs, which not only leverages the dialogue corpus for restructuring missing entity relations but also dynamically selects knowledge based on the dialogue context. Finally, we infuse the dialogue-specific subgraphs to decode the recommendation and responses. We conduct experiments on two benchmark CRSs datasets. Experimental results confirm the effectiveness of our proposed method.

Position modeling plays a critical role in Transformers. In this paper, we focus on length extrapolation, i.e., training on short texts while evaluating longer sequences. We define attention resolution as an indicator of extrapolation. Then we propose two designs to improve the above metric of Transformers. Specifically, we introduce a relative position embedding to explicitly maximize attention resolution. Moreover, we use blockwise causal attention during inference for better resolution. We evaluate different Transformer variants with language modeling. Experimental results show that our model achieves strong performance in both interpolation and extrapolation settings. The code will be available at https://aka.ms/LeX-Transformer.