Aggregation of multi-stage features has been revealed to play a significant role in semantic segmentation. Unlike previous methods employing point-wise summation or concatenation for feature aggregation, this study proposes the Category Feature Transformer (CFT) that explores the flow of category embedding and transformation among multi-stage features through the prevalent multi-head attention mechanism. CFT learns unified feature embeddings for individual semantic categories from high-level features during each aggregation process and dynamically broadcasts them to high-resolution features. Integrating the proposed CFT into a typical feature pyramid structure exhibits superior performance over a broad range of backbone networks. We conduct extensive experiments on popular semantic segmentation benchmarks. Specifically, the proposed CFT obtains a compelling 55.1% mIoU with greatly reduced model parameters and computations on the challenging ADE20K dataset.
Effective and fluent close-proximity human-robot interaction requires understanding how humans get habituated to robots and how robot motion affects human comfort. While prior work has identified humans' preferences over robot motion characteristics and studied their influence on comfort, we are yet to understand how novice first-time robot users get habituated to robots and how robot motion impacts the dynamics of comfort over repeated interactions. To take the first step towards such understanding, we carry out a user study to investigate the connections between robot motion and user comfort and habituation. Specifically, we study the influence of workspace overlap, end-effector speed, and robot motion legibility on overall comfort and its evolution over repeated interactions. Our analyses reveal that workspace overlap, in contrast to speed and legibility, has a significant impact on users' perceived comfort and habituation. In particular, lower workspace overlap leads to users reporting significantly higher overall comfort, lower variations in comfort, and fewer fluctuations in comfort levels during habituation.
Vision transformers have achieved leading performance on various visual tasks yet still suffer from high computational complexity. The situation deteriorates in dense prediction tasks like semantic segmentation, as high-resolution inputs and outputs usually imply more tokens involved in computations. Directly removing the less attentive tokens has been discussed for the image classification task but can not be extended to semantic segmentation since a dense prediction is required for every patch. To this end, this work introduces a Dynamic Token Pruning (DToP) method based on the early exit of tokens for semantic segmentation. Motivated by the coarse-to-fine segmentation process by humans, we naturally split the widely adopted auxiliary-loss-based network architecture into several stages, where each auxiliary block grades every token's difficulty level. We can finalize the prediction of easy tokens in advance without completing the entire forward pass. Moreover, we keep $k$ highest confidence tokens for each semantic category to uphold the representative context information. Thus, computational complexity will change with the difficulty of the input, akin to the way humans do segmentation. Experiments suggest that the proposed DToP architecture reduces on average $20\% - 35\%$ of computational cost for current semantic segmentation methods based on plain vision transformers without accuracy degradation.
The discrimination of instance embeddings plays a vital role in associating instances across time for online video instance segmentation (VIS). Instance embedding learning is directly supervised by the contrastive loss computed upon the contrastive items (CIs), which are sets of anchor/positive/negative embeddings. Recent online VIS methods leverage CIs sourced from one reference frame only, which we argue is insufficient for learning highly discriminative embeddings. Intuitively, a possible strategy to enhance CIs is replicating the inference phase during training. To this end, we propose a simple yet effective training strategy, called Consistent Training for Online VIS (CTVIS), which devotes to aligning the training and inference pipelines in terms of building CIs. Specifically, CTVIS constructs CIs by referring inference the momentum-averaged embedding and the memory bank storage mechanisms, and adding noise to the relevant embeddings. Such an extension allows a reliable comparison between embeddings of current instances and the stable representations of historical instances, thereby conferring an advantage in modeling VIS challenges such as occlusion, re-identification, and deformation. Empirically, CTVIS outstrips the SOTA VIS models by up to +5.0 points on three VIS benchmarks, including YTVIS19 (55.1% AP), YTVIS21 (50.1% AP) and OVIS (35.5% AP). Furthermore, we find that pseudo-videos transformed from images can train robust models surpassing fully-supervised ones.
Current approaches for knowledge distillation in semantic segmentation tend to adopt a holistic approach that treats all spatial locations equally. However, for dense prediction tasks, it is crucial to consider the knowledge representation for different spatial locations in a different manner. Furthermore, edge regions between adjacent categories are highly uncertain due to context information leakage, which is particularly pronounced for compact networks. To address this challenge, this paper proposes a novel approach called boundary-privileged knowledge distillation (BPKD). BPKD distills the knowledge of the teacher model's body and edges separately from the compact student model. Specifically, we employ two distinct loss functions: 1) Edge Loss, which aims to distinguish between ambiguous classes at the pixel level in edge regions. 2) Body Loss, which utilizes shape constraints and selectively attends to the inner-semantic regions. Our experiments demonstrate that the proposed BPKD method provides extensive refinements and aggregation for edge and body regions. Additionally, the method achieves state-of-the-art distillation performance for semantic segmentation on three popular benchmark datasets, highlighting its effectiveness and generalization ability. BPKD shows consistent improvements over various lightweight semantic segmentation structures. The code is available at \url{https://github.com/AkideLiu/BPKD}.
We explore the capability of plain Vision Transformers (ViTs) for semantic segmentation using the encoder-decoder framework and introduce SegViTv2. In our work, we implement the decoder with the global attention mechanism inherent in ViT backbones and propose the lightweight Attention-to-Mask module that effectively converts the global attention map into semantic masks for high-quality segmentation results. Our decoder can outperform the most commonly-used decoder UpperNet in various ViT backbones while consuming only about 5\% of the computational cost. For the encoder, we address the concern of the relatively high computational cost in the ViT-based encoders and propose a Shrunk++ structure that incorporates edge-aware query-based down-sampling (EQD) and query-based up-sampling (QU) modules. The Shrunk++ structure reduces the computational cost of the encoder by up to $50\%$ while maintaining competitive performance. Furthermore, due to the flexibility of our ViT-based architecture, SegVit can be easily extended to semantic segmentation under the setting of continual learning, achieving nearly zero forgetting. Experiments show that our proposed SegViT outperforms recent segmentation methods on three popular benchmarks including ADE20k, COCO-Stuff-10k and PASCAL-Context datasets. The code is available through the following link: \url{https://github.com/zbwxp/SegVit}.
Multi-task visual perception has a wide range of applications in scene understanding such as autonomous driving. In this work, we devise an efficient unified framework to solve multiple common perception tasks, including instance segmentation, semantic segmentation, monocular 3D detection, and depth estimation. Simply sharing the same visual feature representations for these tasks impairs the performance of tasks, while independent task-specific feature extractors lead to parameter redundancy and latency. Thus, we design two feature-merge branches to learn feature basis, which can be useful to, and thus shared by, multiple perception tasks. Then, each task takes the corresponding feature basis as the input of the prediction task head to fulfill a specific task. In particular, one feature merge branch is designed for instance-level recognition the other for dense predictions. To enhance inter-branch communication, the instance branch passes pixel-wise spatial information of each instance to the dense branch using efficient dynamic convolution weighting. Moreover, a simple but effective dynamic routing mechanism is proposed to isolate task-specific features and leverage common properties among tasks. Our proposed framework, termed D2BNet, demonstrates a unique approach to parameter-efficient predictions for multi-task perception. In addition, as tasks benefit from co-training with each other, our solution achieves on par results on partially labeled settings on nuScenes and outperforms previous works for 3D detection and depth estimation on the Cityscapes dataset with full supervision.
Prompt learning has become a popular approach for adapting large vision-language models, such as CLIP, to downstream tasks. Typically, prompt learning relies on a fixed prompt token or an input-conditional token to fit a small amount of data under full supervision. While this paradigm can generalize to a certain range of unseen classes, it may struggle when domain gap increases, such as in fine-grained classification and satellite image segmentation. To address this limitation, we propose Retrieval-enhanced Prompt learning (RePrompt), which introduces retrieval mechanisms to cache the knowledge representations from downstream tasks. we first construct a retrieval database from training examples, or from external examples when available. We then integrate this retrieval-enhanced mechanism into various stages of a simple prompt learning baseline. By referencing similar samples in the training set, the enhanced model is better able to adapt to new tasks with few samples. Our extensive experiments over 15 vision datasets, including 11 downstream tasks with few-shot setting and 4 domain generalization benchmarks, demonstrate that RePrompt achieves considerably improved performance. Our proposed approach provides a promising solution to the challenges faced by prompt learning when domain gap increases. The code and models will be available.
The recent Segment Anything Model (SAM) represents a big leap in scaling up segmentation models, allowing for powerful zero-shot capabilities and flexible prompting. Despite being trained with 1.1 billion masks, SAM's mask prediction quality falls short in many cases, particularly when dealing with objects that have intricate structures. We propose HQ-SAM, equipping SAM with the ability to accurately segment any object, while maintaining SAM's original promptable design, efficiency, and zero-shot generalizability. Our careful design reuses and preserves the pre-trained model weights of SAM, while only introducing minimal additional parameters and computation. We design a learnable High-Quality Output Token, which is injected into SAM's mask decoder and is responsible for predicting the high-quality mask. Instead of only applying it on mask-decoder features, we first fuse them with early and final ViT features for improved mask details. To train our introduced learnable parameters, we compose a dataset of 44K fine-grained masks from several sources. HQ-SAM is only trained on the introduced detaset of 44k masks, which takes only 4 hours on 8 GPUs. We show the efficacy of HQ-SAM in a suite of 9 diverse segmentation datasets across different downstream tasks, where 7 out of them are evaluated in a zero-shot transfer protocol. Our code and models will be released at https://github.com/SysCV/SAM-HQ.
Conversational recommendation systems (CRS) aim to timely and proactively acquire user dynamic preferred attributes through conversations for item recommendation. In each turn of CRS, there naturally have two decision-making processes with different roles that influence each other: 1) director, which is to select the follow-up option (i.e., ask or recommend) that is more effective for reducing the action space and acquiring user preferences; and 2) actor, which is to accordingly choose primitive actions (i.e., asked attribute or recommended item) that satisfy user preferences and give feedback to estimate the effectiveness of the director's option. However, existing methods heavily rely on a unified decision-making module or heuristic rules, while neglecting to distinguish the roles of different decision procedures, as well as the mutual influences between them. To address this, we propose a novel Director-Actor Hierarchical Conversational Recommender (DAHCR), where the director selects the most effective option, followed by the actor accordingly choosing primitive actions that satisfy user preferences. Specifically, we develop a dynamic hypergraph to model user preferences and introduce an intrinsic motivation to train from weak supervision over the director. Finally, to alleviate the bad effect of model bias on the mutual influence between the director and actor, we model the director's option by sampling from a categorical distribution. Extensive experiments demonstrate that DAHCR outperforms state-of-the-art methods.