This paper investigates how to realize better and more efficient embedding learning to tackle the semi-supervised video object segmentation under challenging multi-object scenarios. The state-of-the-art methods learn to decode features with a single positive object and thus have to match and segment each target separately under multi-object scenarios, consuming multiple times computing resources. To solve the problem, we propose an Associating Objects with Transformers (AOT) approach to match and decode multiple objects uniformly. In detail, AOT employs an identification mechanism to associate multiple targets into the same high-dimensional embedding space. Thus, we can simultaneously process the matching and segmentation decoding of multiple objects as efficiently as processing a single object. For sufficiently modeling multi-object association, a Long Short-Term Transformer is designed for constructing hierarchical matching and propagation. We conduct extensive experiments on both multi-object and single-object benchmarks to examine AOT variant networks with different complexities. Particularly, our AOT-L outperforms all the state-of-the-art competitors on three popular benchmarks, i.e., YouTube-VOS (83.7% J&F), DAVIS 2017 (83.0%), and DAVIS 2016 (91.0%), while keeping better multi-object efficiency. Meanwhile, our AOT-T can maintain real-time multi-object speed on above benchmarks. We ranked 1st in the 3rd Large-scale Video Object Segmentation Challenge. The code will be publicly available at https://github.com/z-x-yang/AOT.
Continuity of care is crucial to ensuring positive health outcomes for patients discharged from an inpatient hospital setting, and improved information sharing can help. To share information, caregivers write discharge notes containing action items to share with patients and their future caregivers, but these action items are easily lost due to the lengthiness of the documents. In this work, we describe our creation of a dataset of clinical action items annotated over MIMIC-III, the largest publicly available dataset of real clinical notes. This dataset, which we call CLIP, is annotated by physicians and covers 718 documents representing 100K sentences. We describe the task of extracting the action items from these documents as multi-aspect extractive summarization, with each aspect representing a type of action to be taken. We evaluate several machine learning models on this task, and show that the best models exploit in-domain language model pre-training on 59K unannotated documents, and incorporate context from neighboring sentences. We also propose an approach to pre-training data selection that allows us to explore the trade-off between size and domain-specificity of pre-training datasets for this task.
Few-shot segmentation aims to train a segmentation model that can fast adapt to novel classes with few exemplars. The conventional training paradigm is to learn to make predictions on query images conditioned on the features from support images. Previous methods only utilized the semantic-level prototypes of support images as the conditional information. These methods cannot utilize all pixel-wise support information for the query predictions, which is however critical for the segmentation task. In this paper, we focus on utilizing pixel-wise relationships between support and target images to facilitate the few-shot semantic segmentation task. We design a novel Cycle-Consistent Transformer (CyCTR) module to aggregate pixel-wise support features into query ones. CyCTR performs cross-attention between features from different images, i.e. support and query images. We observe that there may exist unexpected irrelevant pixel-level support features. Directly performing cross-attention may aggregate these features from support to query and bias the query features. Thus, we propose using a novel cycle-consistent attention mechanism to filter out possible harmful support features and encourage query features to attend to the most informative pixels from support images. Experiments on all few-shot segmentation benchmarks demonstrate that our proposed CyCTR leads to remarkable improvement compared to previous state-of-the-art methods. Specifically, on Pascal-$5^i$ and COCO-$20^i$ datasets, we achieve 66.6% and 45.6% mIoU for 5-shot segmentation, outperforming previous state-of-the-art by 4.6% and 7.1% respectively.
Obtaining viewer responses from videos can be useful for creators and streaming platforms to analyze the video performance and improve the future user experience. In this report, we present our method for 2021 Evoked Expression from Videos Challenge. In particular, our model utilizes both audio and image modalities as inputs to predict emotion changes of viewers. To model long-range emotion changes, we use a GRU-based model to predict one sparse signal with 1Hz. We observe that the emotion changes are smooth. Therefore, the final dense prediction is obtained via linear interpolating the signal, which is robust to the prediction fluctuation. Albeit simple, the proposed method has achieved pearson's correlation score of 0.04430 on the final private test set.
Referring video object segmentation (RVOS) aims to segment video objects with the guidance of natural language reference. Previous methods typically tackle RVOS through directly grounding linguistic reference over the image lattice. Such bottom-up strategy fails to explore object-level cues, easily leading to inferior results. In this work, we instead put forward a two-stage, top-down RVOS solution. First, an exhaustive set of object tracklets is constructed by propagating object masks detected from several sampled frames to the entire video. Second, a Transformer-based tracklet-language grounding module is proposed, which models instance-level visual relations and cross-modal interactions simultaneously and efficiently. Our model ranks first place on CVPR2021 Referring Youtube-VOS challenge.
Few-shot segmentation~(FSS) performance has been extensively promoted by introducing episodic training and class-wise prototypes. However, the FSS problem remains challenging due to three limitations: (1) Models are distracted by task-unrelated information; (2) The representation ability of a single prototype is limited; (3) Class-related prototypes ignore the prior knowledge of base classes. We propose the Prior-Enhanced network with Meta-Prototypes to tackle these limitations. The prior-enhanced network leverages the support and query (pseudo-) labels in feature extraction, which guides the model to focus on the task-related features of the foreground objects, and suppress much noise due to the lack of supervised knowledge. Moreover, we introduce multiple meta-prototypes to encode hierarchical features and learn class-agnostic structural information. The hierarchical features help the model highlight the decision boundary and focus on hard pixels, and the structural information learned from base classes is treated as the prior knowledge for novel classes. Experiments show that our method achieves the mean-IoU scores of 60.79% and 41.16% on PASCAL-$5^i$ and COCO-$20^i$, outperforming the state-of-the-art method by 3.49% and 5.64% in the 5-shot setting. Moreover, comparing with 1-shot results, our method promotes 5-shot accuracy by 3.73% and 10.32% on the above two benchmarks. The source code of our method is available at https://github.com/Jarvis73/PEMP.
In general, many dynamic processes are involved with interacting variables, from physical systems to sociological analysis. The interplay of components in the system can give rise to confounding dynamic behavior. Many approaches model temporal sequences holistically ignoring the internal interaction which are impotent in capturing the protogenic actuation. Differently, our goal is to represent a system with a part-whole hierarchy and discover the implied dependencies among intra-system variables: inferring the interactions that possess causal effects on the sub-system behavior with REcurrent partItioned Network (REIN). The proposed architecture consists of (i) a perceptive module that extracts a hierarchical and temporally consistent representation of the observation at multiple levels, (ii) a deductive module for determining the relational connection between neurons at each level, and (iii) a statistical module that can predict the future by conditioning on the temporal distributional estimation. Our model is demonstrated to be effective in identifying the componential interactions with limited observation and stable in long-term future predictions experimented with diverse physical systems.
In this paper, we investigate the task of hallucinating an authentic high-resolution (HR) human face from multiple low-resolution (LR) video snapshots. We propose a pure transformer-based model, dubbed VidFace, to fully exploit the full-range spatio-temporal information and facial structure cues among multiple thumbnails. Specifically, VidFace handles multiple snapshots all at once and harnesses the spatial and temporal information integrally to explore face alignments across all the frames, thus avoiding accumulating alignment errors. Moreover, we design a recurrent position embedding module to equip our transformer with facial priors, which not only effectively regularises the alignment mechanism but also supplants notorious pre-training. Finally, we curate a new large-scale video face hallucination dataset from the public Voxceleb2 benchmark, which challenges prior arts on tackling unaligned and tiny face snapshots. To the best of our knowledge, we are the first attempt to develop a unified transformer-based solver tailored for video-based face hallucination. Extensive experiments on public video face benchmarks show that the proposed method significantly outperforms the state of the arts.
Vehicle search is one basic task for the efficient traffic management in terms of the AI City. Most existing practices focus on the image-based vehicle matching, including vehicle re-identification and vehicle tracking. In this paper, we apply one new modality, i.e., the language description, to search the vehicle of interest and explore the potential of this task in the real-world scenario. The natural language-based vehicle search poses one new challenge of fine-grained understanding of both vision and language modalities. To connect language and vision, we propose to jointly train the state-of-the-art vision models with the transformer-based language model in an end-to-end manner. Except for the network structure design and the training strategy, several optimization objectives are also re-visited in this work. The qualitative and quantitative experiments verify the effectiveness of the proposed method. Our proposed method has achieved the 1st place on the 5th AI City Challenge, yielding competitive performance 18.69% MRR accuracy on the private test set. We hope this work can pave the way for the future study on using language description effectively and efficiently for real-world vehicle retrieval systems. The code will be available at https://github.com/ShuaiBai623/AIC2021-T5-CLV.
Contemporary data-driven methods are typically fed with full supervision on large-scale datasets which limits their applicability. However, in the actual systems with limitations such as measurement error and data acquisition problems, people usually obtain incomplete data. Although data completion has attracted wide attention, the underlying data pattern and relativity are still under-developed. Currently, the family of latent variable models allows learning deep latent variables over observed variables by fitting the marginal distribution. As far as we know, current methods fail to perceive the data relativity under partial observation. Aiming at modeling incomplete data, this work uses relational inference to fill in the incomplete data. Specifically, we expect to approximate the real joint distribution over the partial observation and latent variables, thus infer the unseen targets respectively. To this end, we propose Omni-Relational Network (OR-Net) to model the pointwise relativity in two aspects: (i) On one hand, the inner relationship is built among the context points in the partial observation; (ii) On the other hand, the unseen targets are inferred by learning the cross-relationship with the observed data points. It is further discovered that the proposed method can be generalized to different scenarios regardless of whether the physical structure can be observed or not. It is demonstrated that the proposed OR-Net can be well generalized for data completion tasks of various modalities, including function regression, image completion on MNIST and CelebA datasets, and also sequential motion generation conditioned on the observed poses.