Today's scene graph generation (SGG) models typically require abundant manual annotations to learn new predicate types. Thus, it is difficult to apply them to real-world applications with a long-tailed distribution of predicates. In this paper, we focus on a new promising task of SGG: few-shot SGG (FSSGG). FSSGG encourages models to be able to quickly transfer previous knowledge and recognize novel predicates well with only a few examples. Although many advanced approaches have achieved great success on few-shot learning (FSL) tasks, straightforwardly extending them into FSSGG is not applicable due to two intrinsic characteristics of predicate concepts: 1) Each predicate category commonly has multiple semantic meanings under different contexts. 2) The visual appearance of relation triplets with the same predicate differs greatly under different subject-object pairs. Both issues make it hard to model conventional latent representations for predicate categories with state-of-the-art FSL methods. To this end, we propose a novel Decomposed Prototype Learning (DPL). Specifically, we first construct a decomposable prototype space to capture intrinsic visual patterns of subjects and objects for predicates, and enhance their feature representations with these decomposed prototypes. Then, we devise an intelligent metric learner to assign adaptive weights to each support sample by considering the relevance of their subject-object pairs. We further re-split the VG dataset and compare DPL with various FSL methods to benchmark this task. Extensive results show that DPL achieves excellent performance in both base and novel categories.
Intelligent vehicles (IVs) have attracted wide attention thanks to the augmented convenience, safety advantages, and potential commercial value. Although a few of autonomous driving unicorns assert that IVs will be commercially deployable by 2025, their deployment is still restricted to small-scale validation due to various issues, among which safety, reliability, and generalization of planning methods are prominent concerns. Precise computation of control commands or trajectories by planning methods remains a prerequisite for IVs, owing to perceptual imperfections under complex environments, which pose an obstacle to the successful commercialization of IVs. This paper aims to review state-of-the-art planning methods, including pipeline planning and end-to-end planning methods. In terms of pipeline methods, a survey of selecting algorithms is provided along with a discussion of the expansion and optimization mechanisms, whereas in end-to-end methods, the training approaches and verification scenarios of driving tasks are points of concern. Experimental platforms are reviewed to facilitate readers in selecting suitable training and validation methods. Finally, the current challenges and future directions are discussed. The side-by-side comparison presented in this survey helps to gain insights into the strengths and limitations of the reviewed methods, which also assists with system-level design choices.
While features of different scales are perceptually important to visual inputs, existing vision transformers do not yet take advantage of them explicitly. To this end, we first propose a cross-scale vision transformer, CrossFormer. It introduces a cross-scale embedding layer (CEL) and a long-short distance attention (LSDA). On the one hand, CEL blends each token with multiple patches of different scales, providing the self-attention module itself with cross-scale features. On the other hand, LSDA splits the self-attention module into a short-distance one and a long-distance counterpart, which not only reduces the computational burden but also keeps both small-scale and large-scale features in the tokens. Moreover, through experiments on CrossFormer, we observe another two issues that affect vision transformers' performance, i.e. the enlarging self-attention maps and amplitude explosion. Thus, we further propose a progressive group size (PGS) paradigm and an amplitude cooling layer (ACL) to alleviate the two issues, respectively. The CrossFormer incorporating with PGS and ACL is called CrossFormer++. Extensive experiments show that CrossFormer++ outperforms the other vision transformers on image classification, object detection, instance segmentation, and semantic segmentation tasks. The code will be available at: https://github.com/cheerss/CrossFormer.
Controllable Image Captioning (CIC) -- generating natural language descriptions about images under the guidance of given control signals -- is one of the most promising directions towards next-generation captioning systems. Till now, various kinds of control signals for CIC have been proposed, ranging from content-related control to structure-related control. However, due to the format and target gaps of different control signals, all existing CIC works (or architectures) only focus on one certain control signal, and overlook the human-like combinatorial ability. By ``combinatorial", we mean that our humans can easily meet multiple needs (or constraints) simultaneously when generating descriptions. To this end, we propose a novel prompt-based framework for CIC by learning Combinatorial Prompts, dubbed as ComPro. Specifically, we directly utilize a pretrained language model GPT-2 as our language model, which can help to bridge the gap between different signal-specific CIC architectures. Then, we reformulate the CIC as a prompt-guide sentence generation problem, and propose a new lightweight prompt generation network to generate the combinatorial prompts for different kinds of control signals. For different control signals, we further design a new mask attention mechanism to realize the prompt-based CIC. Due to its simplicity, our ComPro can easily be extended to more complex combined control signals by concatenating these prompts. Extensive experiments on two prevalent CIC benchmarks have verified the effectiveness and efficiency of our ComPro on both single and combined control signals.
Prompt tuning with large-scale pretrained vision-language models empowers open-vocabulary predictions trained on limited base categories, e.g., object classification and detection. In this paper, we propose compositional prompt tuning with motion cues: an extended prompt tuning paradigm for compositional predictions of video data. In particular, we present Relation Prompt (RePro) for Open-vocabulary Video Visual Relation Detection (Open-VidVRD), where conventional prompt tuning is easily biased to certain subject-object combinations and motion patterns. To this end, RePro addresses the two technical challenges of Open-VidVRD: 1) the prompt tokens should respect the two different semantic roles of subject and object, and 2) the tuning should account for the diverse spatio-temporal motion patterns of the subject-object compositions. Without bells and whistles, our RePro achieves a new state-of-the-art performance on two VidVRD benchmarks of not only the base training object and predicate categories, but also the unseen ones. Extensive ablations also demonstrate the effectiveness of the proposed compositional and multi-mode design of prompts. Code is available at https://github.com/Dawn-LX/OpenVoc-VidVRD.
Video representation learning has been successful in video-text pre-training for zero-shot transfer, where each sentence is trained to be close to the paired video clips in a common feature space. For long videos, given a paragraph of description where the sentences describe different segments of the video, by matching all sentence-clip pairs, the paragraph and the full video are aligned implicitly. However, such unit-level similarity measure may ignore the global temporal context over a long time span, which inevitably limits the generalization ability. In this paper, we propose a contrastive learning framework TempCLR to compare the full video and the paragraph explicitly. As the video/paragraph is formulated as a sequence of clips/sentences, under the constraint of their temporal order, we use dynamic time warping to compute the minimum cumulative cost over sentence-clip pairs as the sequence-level distance. To explore the temporal dynamics, we break the consistency of temporal order by shuffling the video clips or sentences according to the temporal granularity. In this way, we obtain the representations for clips/sentences, which perceive the temporal information and thus facilitate the sequence alignment. In addition to pre-training on the video and paragraph, our approach can also generalize on the matching between different video instances. We evaluate our approach on video retrieval, action step localization, and few-shot action recognition, and achieve consistent performance gain over all three tasks. Detailed ablation studies are provided to justify the approach design.
Given an untrimmed video and natural language query, video sentence grounding aims to localize the target temporal moment in the video. Existing methods mainly tackle this task by matching and aligning semantics of the descriptive sentence and video segments on a single temporal resolution, while neglecting the temporal consistency of video content in different resolutions. In this work, we propose a novel multi-resolution temporal video sentence grounding network: MRTNet, which consists of a multi-modal feature encoder, a Multi-Resolution Temporal (MRT) module, and a predictor module. MRT module is an encoder-decoder network, and output features in the decoder part are in conjunction with Transformers to predict the final start and end timestamps. Particularly, our MRT module is hot-pluggable, which means it can be seamlessly incorporated into any anchor-free models. Besides, we utilize a hybrid loss to supervise cross-modal features in MRT module for more accurate grounding in three scales: frame-level, clip-level and sequence-level. Extensive experiments on three prevalent datasets have shown the effectiveness of MRTNet.
As ride-hailing services become increasingly popular, being able to accurately predict demand for such services can help operators efficiently allocate drivers to customers, and reduce idle time, improve congestion, and enhance the passenger experience. This paper proposes UberNet, a deep learning Convolutional Neural Network for short-term prediction of demand for ride-hailing services. UberNet empploys a multivariate framework that utilises a number of temporal and spatial features that have been found in the literature to explain demand for ride-hailing services. The proposed model includes two sub-networks that aim to encode the source series of various features and decode the predicting series, respectively. To assess the performance and effectiveness of UberNet, we use 9 months of Uber pickup data in 2014 and 28 spatial and temporal features from New York City. By comparing the performance of UberNet with several other approaches, we show that the prediction quality of the model is highly competitive. Further, Ubernet's prediction performance is better when using economic, social and built environment features. This suggests that Ubernet is more naturally suited to including complex motivators in making real-time passenger demand predictions for ride-hailing services.
Person Search aims to simultaneously localize and recognize a target person from realistic and uncropped gallery images. One major challenge of person search comes from the contradictory goals of the two sub-tasks, i.e., person detection focuses on finding the commonness of all persons so as to distinguish persons from the background, while person re-identification (re-ID) focuses on the differences among different persons. In this paper, we propose a novel Sequential Transformer (SeqTR) for end-to-end person search to deal with this challenge. Our SeqTR contains a detection transformer and a novel re-ID transformer that sequentially addresses detection and re-ID tasks. The re-ID transformer comprises the self-attention layer that utilizes contextual information and the cross-attention layer that learns local fine-grained discriminative features of the human body. Moreover, the re-ID transformer is shared and supervised by multi-scale features to improve the robustness of learned person representations. Extensive experiments on two widely-used person search benchmarks, CUHK-SYSU and PRW, show that our proposed SeqTR not only outperforms all existing person search methods with a 59.3% mAP on PRW but also achieves comparable performance to the state-of-the-art results with an mAP of 94.8% on CUHK-SYSU.
Mural image inpainting refers to repairing the damage or missing areas in a mural image to restore the visual appearance. Most existing image-inpainting methods tend to take a target image as the only input and directly repair the damage to generate a visually plausible result. These methods obtain high performance in restoration or completion of some specific objects, e.g., human face, fabric texture, and printed texts, etc., however, are not suitable for repairing murals with varied subjects, especially for murals with large damaged areas. Moreover, due to the discrete colors in paints, mural inpainting may suffer from apparent color bias as compared to natural image inpainting. To this end, in this paper, we propose a line drawing guided progressive mural inpainting method. It divides the inpainting process into two steps: structure reconstruction and color correction, executed by a structure reconstruction network (SRN) and a color correction network (CCN), respectively. In the structure reconstruction, line drawings are used by SRN as a guarantee for large-scale content authenticity and structural stability. In the color correction, CCN operates a local color adjustment for missing pixels which reduces the negative effects of color bias and edge jumping. The proposed approach is evaluated against the current state-of-the-art image inpainting methods. Qualitative and quantitative results demonstrate the superiority of the proposed method in mural image inpainting. The codes and data are available at {https://github.com/qinnzou/mural-image-inpainting}.