12 April 2024 in IEEE Transactions on Pattern Analysis and Machine Intelligence
3D dense captioning requires a model to translate its understanding of an input 3D scene into several captions associated with different object regions. Existing methods adopt a sophisticated “detect-then-describe” pipeline, which builds explicit relation modules upon a 3D detector with numerous hand-crafted components. While these methods have achieved initial success, the cascade pipeline tends to accumulate errors because of duplicated and inaccurate box estimations and messy 3D scenes. In this paper, we first propose Vote2Cap-DETR, a simple-yet-effective transformer framework that decouples the decoding process of caption generation and object localization through parallel decoding. Moreover, we argue that object localization and description generation require different levels of scene understanding, which could be challenging for a shared set of queries to capture. To this end, we propose an advanced version, Vote2Cap-DETR++, which decouples the queries into localization and caption queries to capture task-specific features. Additionally, we introduce the iterative spatial refinement strategy to vote queries for faster convergence and better localization performance. We also insert additional spatial information to the caption head for more accurate descriptions. Without bells and whistles, extensive experiments on two commonly used datasets, ScanRefer and Nr3D, demonstrate Vote2Cap-DETR and Vote2Cap-DETR++ surpass conventional “detect-then-describe” methods by a large margin.
31 December 2019 in IEEE Transactions on Cybernetics
Single-image dehazing has been an important topic given the commonly occurred image degradation caused by adverse atmosphere aerosols. The key to haze removal relies on an accurate estimation of global air-light and the transmission map. Most existing methods estimate these two parameters using separate pipelines which reduces the efficiency and accumulates errors, thus leading to a suboptimal approximation, hurting the model interpretability, and degrading the performance. To address these issues, this article introduces a novel generative adversarial network (GAN) for single-image dehazing. The network consists of a novel compositional generator and a novel deeply supervised discriminator. The compositional generator is a densely connected network, which combines fine-scale and coarse-scale information. Benefiting from the new generator, our method can directly learn the physical parameters from data and recover clean images from hazy ones in an end-to-end manner. The proposed discriminator is deeply supervised, which enforces that the output of the generator to look similar to the clean images from low-level details to high-level structures. To the best of our knowledge, this is the first end-to-end generative adversarial model for image dehazing, which simultaneously outputs clean images, transmission maps, and air-lights. Extensive experiments show that our method remarkably outperforms the state-of-the-art methods. Furthermore, to facilitate future research, we create the HazeCOCO dataset which is currently the largest dataset for single-image dehazing.
14 February 2018 in IEEE Transactions on Image Processing
In this paper, we propose YoTube-a novel deep learning framework for generating action proposals in untrimmed videos, where each action proposal corresponds to a spatial-temporal tube that potentially locates one human action. Most of the existing works generate proposals by clustering low-level features or linking image proposals, which ignore the interplay between long-term temporal context and short-term cues. Different from these works, our method considers the interplay by designing a new recurrent YoTube detector and static YoTube detector. The recurrent YoTube detector sequentially regresses candidate bounding boxes using Recurrent Neural Network learned long-term temporal contexts. The static YoTube detector produces bounding boxes using rich appearance cues in every single frame. To fully exploit the complementary appearance, motion, and temporal context, we train the recurrent and static detector using RGB (Color) and flow information. Moreover, we fuse the corresponding outputs of the detectors to produce accurate and robust proposal boxes and obtain the final action proposals by linking the proposal boxes using dynamic programming with a novel path trimming method. Benefiting from the pipeline of our method, the untrimmed video could be effectively and efficiently handled. Extensive experiments on the challenging UCF-101, UCF-Sports, and JHMDB datasets show superior performance of the proposed method compared with the state of the arts.
28 September 2018 in IEEE Transactions on Industrial Electronics
Dimension reduction is a fundamental task of machine learning and computer vision, which is widely used in a variety of industrial applications. Over past decades, a lot of unsupervised and supervised algorithms have been proposed. However, few of them can automatically determine the feature dimension that could be adaptive to different data distributions. To obtain a good performance, it is popular to seek the optimal dimension by exhaustively enumerating some possible values. Clearly, such a scheme is ad-hoc and computational extensive. Therefore, a method which can automatically estimate the feature dimension in an efficient and principled manner is of significant practical and theoretical value. In this paper, we propose a novel supervised subspace learning method called multiple marginal fisher analysis (MMFA), which can automatically estimate the feature dimension. By maxing the inter-class separability among marginal points while minimizing within-class scatter, MMFA obtains low-dimensional representations with outstanding discriminative properties. Extensive experiments show that MMFA not only outperforms other algorithms on clean data but also show robustness on corrupted and disguised data.
3 February 2021 in IEEE Transactions on Pattern Analysis and Machine Intelligence
Imbalanced data distribution in crowd counting datasets leads to severe under-estimation and over-estimation problems, which has been less investigated in existing works. In this paper, we tackle this challenging problem by proposing a simple but effective locality-based learning paradigm to produce generalizable features by alleviating sample bias. Our proposed method is locality-aware in two aspects. First, we introduce a locality-aware data partition (LADP) approach to group the training data into different bins via locality-sensitive hashing. As a result, a more balanced data batch is then constructed by LADP. To further reduce the training bias and enhance the collaboration with LADP, a new data augmentation method called locality-aware data augmentation (LADA) is proposed where the image patches are adaptively augmented based on the loss. The proposed method is independent of the backbone network architectures, and thus could be smoothly integrated with most existing deep crowd counting approaches in an end-to-end paradigm to boost their performance. We also demonstrate the versatility of the proposed method by applying it for adversarial defense. Extensive experiments verify the superiority of the proposed method over the state of the arts.
7 October 2019 in IEEE Transactions on Industrial Informatics
Night-time pedestrian retrieval is a cross-modality retrieval task of retrieving person images between day-time visible images and night-time thermal images. It is a very challenging problem due to modality difference, camera variations and person variations, but it plays an important role in night-time video surveillance. Existing cross-modality retrieval usually focuses on learning modality sharable feature representations to bridge the modality gap. In this paper, we propose to utilize auxiliary information to improve retrieval performance, which consistently improves the performance with different baseline loss functions. Our auxiliary information contains two major parts: cross-modality feature distribution and contextual information. The former aligns the cross-modality feature distributions between two modalities to improve the performance, and the latter optimizes the cross-modality distance measurement with the contextual information. We also demonstrate that abundant annotated visible pedestrian images, which are easily accessible, helps to improve the cross-modality pedestrian retrieval as well. The proposed method is featured in two aspects: the auxiliary information does not need additional human intervention or annotation; it learns discriminative feature representations in an end-to-end deep learning manner. Extensive experiments on two cross-modality pedestrian retrieval datasets demonstrate the superiority of the proposed method, achieving much better performance than the state-of-the-arts.