An image captioning model flexibly switching its language pattern, e.g., descriptiveness and length, should be useful since it can be applied to diverse applications. However, despite the dramatic improvement in generative vision-language models, fine-grained control over the properties of generated captions is not easy due to two reasons: (i) existing models are not given the properties as a condition during training and (ii) existing models cannot smoothly transition its language pattern from one state to the other. Given this challenge, we propose a new approach, CaptionSmiths, to acquire a single captioning model that can handle diverse language patterns. First, our approach quantifies three properties of each caption, length, descriptiveness, and uniqueness of a word, as continuous scalar values, without human annotation. Given the values, we represent the conditioning via interpolation between two endpoint vectors corresponding to the extreme states, e.g., one for a very short caption and one for a very long caption. Empirical results demonstrate that the resulting model can smoothly change the properties of the output captions and show higher lexical alignment than baselines. For instance, CaptionSmiths reduces the error in controlling caption length by 506\% despite better lexical alignment. Code will be available on https://github.com/omron-sinicx/captionsmiths.

A common practice in metric learning is to train and test an embedding model for each dataset. This dataset-specific approach fails to simulate real-world scenarios that involve multiple heterogeneous distributions of data. In this regard, we explore a new metric learning paradigm, called Uni-fied Metric Learning (UML), which learns a unified dis-tance metric capable of capturing relations across multi-ple data distributions. UML presents new challenges, such as imbalanced data distribution and bias towards dom-inant distributions. These issues cause standard metric learning methods to fail in learning a unified metric. To address these challenges, we propose Parameter-efficient Unified Metric leArning (PUMA), which consists of a pre-trained frozen model and two additional modules, stochas-tic adapter and prompt pool. These modules enable to capture dataset-specific knowledge while avoiding bias to-wards dominant distributions. Additionally, we compile a new unified metric learning benchmark with a total of 8 different datasets. PUMA outperforms the state-of-the-art dataset-specific models while using about 69 times fewer trainable parameters.

Weakly-Supervised Scene Graph Generation (WSSGG) research has recently emerged as an alternative to the fully-supervised approach that heavily relies on costly annotations. In this regard, studies on WSSGG have utilized image captions to obtain unlocalized triplets while primarily focusing on grounding the unlocalized triplets over image regions. However, they have overlooked the two issues involved in the triplet formation process from the captions: 1) Semantic over-simplification issue arises when extracting triplets from captions, where fine- grained predicates in captions are undesirably converted into coarse-grained predi-cates, resulting in a long-tailed predicate distribution, and 2) Low-density scene graph issue arises when aligning the triplets in the caption with entity/predicate classes of interest, where many triplets are discarded and not used in training, leading to insufficient supervision. To tackle the two issues, we propose a new approach, i.e., Large Language Modelfor weakly-supervised SGG (LLM4SGG), where we mitigate the two issues by leveraging the LLM' sin-depth understanding of language and reasoning ability during the extraction of triplets from captions and alignment of entity/predicate classes with target data. To further engage the LLM in these processes, we adopt the idea of Chain-of-Thought and the in-context few-shot learning strategy. To validate the effectiveness of LLM4SGG, we conduct extensive experiments on Visual Genome and GQA datasets, showing significant improvements in both Recall@K and mean Recall@K compared to the state-of-the-art WSSGG methods. A further appeal is that LLM4SGG is data-efficient, enabling effective model training with a small amount of training images. Our code is available on https://github.com/rlqjall07/torch-LLM4SGG

Scene graph generation (SGG) models have suffered from inherent problems regarding the benchmark datasets such as the long-tailed predicate distribution and missing annotation problems. In this work, we aim to alleviate the long-tailed problem of SGG by utilizing unannotated triplets. To this end, we introduce a **S**elf-**T**raining framework for **SGG** **(ST-SGG)** that assigns pseudo-labels for unannotated triplets based on which the SGG models are trained. While there has been significant progress in self-training for image recognition, designing a self-training framework for the SGG task is more challenging due to its inherent nature such as the semantic ambiguity and the long-tailed distribution of predicate classes. Hence, we propose a novel pseudo-labeling technique for SGG, called **C**lass-specific **A**daptive **T**hresholding with **M**omentum **(CATM)**, which is a model-agnostic framework that can be applied to any existing SGG models. Furthermore, we devise a graph structure learner (GSL) that is beneficial when adopting our proposed self-training framework to the state-of-the-art message-passing neural network (MPNN)-based SGG models. Our extensive experiments verify the effectiveness of ST-SGG on various SGG models, particularly in enhancing the performance on fine-grained predicate classes.

Recent advances in 3D object detection leveraging multi-view cameras have demonstrated their practical and economical value in various challenging vision tasks. However, typical supervised learning approaches face challenges in achieving satisfactory adaptation toward unseen and unlabeled target datasets (i.e., direct transfer) due to the inevitable geometric misalignment between the source and target domains. In practice, we also encounter constraints on resources for training models and collecting annotations for the successful deployment of 3D object detectors. In this paper, we propose Unified Domain Generalization and Adaptation (UDGA), a practical solution to mitigate those drawbacks. We first propose Multi-view Overlap Depth Constraint that leverages the strong association between multi-view, significantly alleviating geometric gaps due to perspective view changes. Then, we present a Label-Efficient Domain Adaptation approach to handle unfamiliar targets with significantly fewer amounts of labels (i.e., 1$\%$ and 5$\%)$, while preserving well-defined source knowledge for training efficiency. Overall, UDGA framework enables stable detection performance in both source and target domains, effectively bridging inevitable domain gaps, while demanding fewer annotations. We demonstrate the robustness of UDGA with large-scale benchmarks: nuScenes, Lyft, and Waymo, where our framework outperforms the current state-of-the-art methods.

Vision Transformers (ViTs) have recently become the state-of-the-art across many computer vision tasks. In contrast to convolutional networks (CNNs), ViTs enable global information sharing even within shallow layers of a network, i.e., among high-resolution features. However, this perk was later overlooked with the success of pyramid architectures such as Swin Transformer, which show better performance-complexity trade-offs. In this paper, we present a simple and efficient add-on component (termed GrafT) that considers global dependencies and multi-scale information throughout the network, in both high- and low-resolution features alike. It has the flexibility of branching out at arbitrary depths and shares most of the parameters and computations of the backbone. GrafT shows consistent gains over various well-known models which includes both hybrid and pure Transformer types, both homogeneous and pyramid structures, and various self-attention methods. In particular, it largely benefits mobile-size models by providing high-level semantics. On the ImageNet-1k dataset, GrafT delivers +3.9%, +1.4%, and +1.9% top-1 accuracy improvement to DeiT-T, Swin-T, and MobViTXXS, respectively. Our code and models are at https://github.com/jongwoopark7978/Grafting-Vision-Transformer.

Recently, large-scale pre-trained Vision-and-Language (VL) foundation models have demonstrated remarkable capabilities in many zero-shot downstream tasks, achieving competitive results for recognizing objects defined by as little as short text prompts. However, it has also been shown that VL models are still brittle in Structured VL Concept (SVLC) reasoning, such as the ability to recognize object attributes, states, and inter-object relations. This leads to reasoning mistakes, which need to be corrected as they occur by teaching VL models the missing SVLC skills; often this must be done using private data where the issue was found, which naturally leads to a data-free continual (no task-id) VL learning setting. In this work, we introduce the first Continual Data-Free Structured VL Concepts Learning (ConStruct-VL) benchmark11Our code is publicly available at https://github.com/jamessealesmith/ConStruct-VL and show it is challenging for many existing data-free CL strategies. We, therefore, propose a data-free method comprised of a new approach of Adversarial Pseudo-Replay (APR) which generates adversarial reminders of past tasks from past task models. To use this method efficiently, we also propose a continual parameter-efficient Layered-LoRA (LaLo) neural architecture allowing no-memory-cost access to all past models at train time. We show this approach outperforms all data-free methods by as much as ~ 7% while even matching some levels of experience-replay (prohibitive for applications where data-privacy must be preserved).

Computer vision models suffer from a phenomenon known as catastrophic forgetting when learning novel concepts from continuously shifting training data. Typical solutions for this continual learning problem require extensive rehearsal of previously seen data, which increases memory costs and may violate data privacy. Recently, the emergence of large-scale pre-trained vision transformer models has enabled prompting approaches as an alternative to data-rehearsal. These approaches rely on a key-query mechanism to generate prompts and have been found to be highly resistant to catastrophic forgetting in the well-established rehearsal-free continual learning setting. However, the key mechanism of these methods is not trained end-to-end with the task sequence. Our experiments show that this leads to a reduction in their plasticity, hence sacrificing new task accuracy, and inability to benefit from expanded parameter capacity. We instead propose to learn a set of prompt components which are assembled with input-conditioned weights to produce input-conditioned prompts, resulting in a novel attention-based end-to-end key-query scheme. Our experiments show that we outperform the current SOTA method DualPrompt on established benchmarks by as much as 4.5% in average final accuracy. We also outperform the state of art by as much as 4.4% accuracy on a continual learning benchmark which contains both class-incremental and domain-incremental task shifts, corresponding to many practical settings. Our code is available at https://github.com/GT-RIPL/CODA-Prompt

Large-scale pre-trained Vision & Language (VL) models have shown remarkable performance in many applications, enabling replacing a fixed set of supported classes with zero-shot open vocabulary reasoning over (almost arbitrary) natural language prompts. However, recent works have uncovered a fundamental weakness of these models. For example, their difficulty to understand Visual Language Concepts (VLC) that go ‘beyond nouns’ such as the meaning of non-object words (e.g., attributes, actions, relations, states, etc.), or difficulty in performing compositional reasoning such as understanding the significance of the order of the words in a sentence. In this work, we investigate to which extent purely synthetic data could be leveraged to teach these models to overcome such shortcomings without compromising their zero-shot capabilities. We contribute Synthetic Visual Concepts (SyViC) - a million-scale synthetic dataset and data generation codebase allowing to generate additional suitable data to improve VLC understanding and compositional reasoning of VL models. Additionally, we propose a general VL finetuning strategy for effectively leveraging SyViC towards achieving these improvements. Our extensive experiments and ablations on VL-Checklist, Winoground, and ARO benchmarks demonstrate that it is possible to adapt strong pre-trained VL models with synthetic data significantly enhancing their VLC understanding (e.g. by 9.9% on ARO and 4.3% on VL-Checklist) with under 1% drop in their zero-shot accuracy.

While transformers have greatly boosted performance in semantic segmentation, domain adaptive transformers are not yet well explored. We identify that the domain gap can cause discrepancies in self-attention. Due to this gap, the transformer attends to spurious regions or pixels, which deteriorates accuracy on the target domain. We propose Cross-Domain Attention Consistency (CDAC), to perform adaptation on attention maps using cross-domain attention layers that share features between source and target domains. Specifically, we impose consistency between predictions from cross-domain attention and self-attention modules to encourage similar distributions across domains in both the attention and output of the model, i.e., attention-level and output-level alignment. We also enforce consistency in attention maps between different augmented views to further strengthen the attention-based alignment. Combining these two components, CDAC mitigates the discrepancy in attention maps across domains and further boosts the performance of the transformer under unsupervised domain adaptation settings. Our method is evaluated on various widely used benchmarks and outperforms the state-of-the-art baselines, including GTAV-to-Cityscapes by 1.3 and 1.5 percent point (pp) and Synthia-to-Cityscapes by 0.6 pp and 2.9 pp when combining with two competitive Transformer-based backbones, respectively. Our code will be publicly available at https://github.com/wangkaihong/CDAC.

Vision and Language (VL) models offer an effective method for aligning representation spaces of images and text allowing for numerous applications such as cross-modal retrieval, visual and multi-hop question answering, captioning, and many more. However, the aligned image-text spaces learned by all the popular VL models are still suffering from the so-called 'object bias' - their representations behave as 'bags of nouns' mostly ignoring or downsizing the attributes, relations, and states of objects described/appearing in texts/images. Although some great attempts at fixing these `compositional reasoning' issues were proposed in the recent literature, the problem is still far from being solved. In this paper, we uncover two factors limiting the VL models' compositional reasoning performance. These two factors are properties of the paired VL dataset used for finetuning (or pre-training) the VL model: (i) the caption quality, or in other words 'image-alignment', of the texts; and (ii) the 'density' of the captions in the sense of mentioning all the details appearing on the image. We propose a fine-tuning approach for automatically treating these factors on a standard collection of paired VL data (CC3M). Applied to CLIP, we demonstrate its significant compositional reasoning performance increase of up to $\sim27$\% over the base model, up to $\sim20$\% over the strongest baseline, and by $6.7$\% on average. Our code is provided in the Supplementary and would be released upon acceptance.

Pre-training on massive video datasets has become essential to achieve high action recognition performance on smaller downstream datasets. However, most large-scale video datasets contain images of people and hence are accompanied with issues related to privacy, ethics, and data protection, often preventing them from being publicly shared for reproducible research. Existing work has attempted to alleviate these problems by blurring faces, downsampling videos, or training on synthetic data. On the other hand, analysis on the {\em transferability} of privacy-preserving pre-trained models to downstream tasks has been limited. In this work, we study this problem by first asking the question: can we pre-train models for human action recognition with data that does not include real humans? To this end, we present, for the first time, a benchmark that leverages real-world videos with {\em humans removed} and synthetic data containing virtual humans to pre-train a model. We then evaluate the transferability of the representation learned on this data to a diverse set of downstream action recognition benchmarks. Furthermore, we propose a novel pre-training strategy, called Privacy-Preserving MAE-Align, to effectively combine synthetic data and human-removed real data. Our approach outperforms previous baselines by up to 5\% and closes the performance gap between human and no-human action recognition representations on downstream tasks, for both linear probing and fine-tuning. Our benchmark, code, and models are available at https://github.com/howardzh01/PPMA.

We present a two-stage pre-training approach that improves the generalization ability of standard single-domain pre-training. While standard pre-training on a single large dataset (such as ImageNet) can provide a good initial representation for transfer learning tasks, this approach may result in biased representations that impact the success of learning with new multi-domain data (e.g., different artistic styles) via methods like domain adaptation. We propose a novel pre-training approach called Cross-Domain Self-supervision (CDS), which directly employs unlabeled multi-domain data for downstream domain transfer tasks. Our approach uses self-supervision not only within a single domain but also across domains. In-domain instance discrimination is used to learn discriminative features on new data in a domain-adaptive manner, while cross-domain matching is used to learn domain-invariant features. We apply our method as a second pre-training step (after ImageNet pre-training), resulting in a significant target accuracy boost to diverse domain transfer tasks compared to standard one-stage pre-training.

Learning transferable and domain adaptive feature representations from videos is important for video-relevant tasks such as action recognition. Existing video domain adaptation methods mainly rely on adversarial feature alignment, which has been derived from the RGB image space. However, video data is usually associated with multi-modal information, e.g., RGB and optical flow, and thus it remains a challenge to design a better method that considers the cross-modal inputs under the cross-domain adaptation setting. To this end, we propose a unified framework for video domain adaptation, which simultaneously regularizes cross-modal and cross-domain feature representations. Specifically, we treat each modality in a domain as a view and leverage the contrastive learning technique with properly designed sampling strategies. As a result, our objectives regularize feature spaces, which originally lack the connection across modalities or have less alignment across domains. We conduct experiments on domain adaptive action recognition benchmark datasets, i.e., UCF, HMDB, and EPIC-Kitchens, and demonstrate the effectiveness of our components against state-of-the-art algorithms.

Unsupervised domain adaptation (UDA) methods can dramatically improve generalization on unlabeled target domains. However, optimal hyper-parameter selection is critical to achieving high accuracy and avoiding negative transfer. Supervised hyper-parameter validation is not possible without labeled target data, which raises the question: How can we validate unsupervised adaptation techniques in a realistic way? We first empirically analyze existing criteria and demonstrate that they are not very effective for tuning hyper-parameters. Intuitively, a well-trained source classifier should embed target samples of the same class nearby, forming dense neighborhoods in feature space. Based on this assumption, we propose a novel unsupervised validation criterion that measures the density of soft neighborhoods by computing the entropy of the similarity distribution between points. Our criterion is simpler than competing validation methods, yet more effective; it can tune hyper-parameters and the number of training iterations in both image classification and semantic segmentation models. The code used for the paper will be available at https://github.com/VisionLearningGroup/SND.

Semi-supervised learning (SSL) is an effective means to leverage unlabeled data to improve a model’s performance. Typical SSL methods like FixMatch assume that labeled and unlabeled data share the same label space. However, in practice, unlabeled data can contain categories unseen in the labeled set, i.e., outliers, which can significantly harm the performance of SSL algorithms. To address this problem, we propose a novel Open-set Semi-Supervised Learning (OSSL) approach called OpenMatch.Learning representations of inliers while rejecting outliers is essential for the success of OSSL. To this end, OpenMatch unifies FixMatch with novelty detection based on one-vs-all (OVA) classifiers. The OVA-classifier outputs the confidence score of a sample being an inlier, providing a threshold to detect outliers. Another key contribution is an open-set soft-consistency regularization loss, which enhances the smoothness of the OVA-classifier with respect to input transformations and greatly improves outlier detection. \ours achieves state-of-the-art performance on three datasets, and even outperforms a fully supervised model in detecting outliers unseen in unlabeled data on CIFAR10. The code is available at \url{https://github.com/VisionLearningGroup/OP_Match}.

Existing vision-language methods typically support two languages at a time at most. In this paper, we present a modular approach which can easily be incorporated into existing vision-language methods in order to support many languages. We accomplish this by learning a single shared Multimodal Universal Language Embedding (MULE) which has been visually-semantically aligned across all languages. Then we learn to relate MULE to visual data as if it were a single language. Our method is not architecture specific, unlike prior work which typically learned separate branches for each language, enabling our approach to easily be adapted to many vision-language methods and tasks. Since MULE learns a single language branch in the multimodal model, we can also scale to support many languages, and languages with fewer annotations can take advantage of the good representation learned from other (more abundant) language data. We demonstrate the effectiveness of our embeddings on the bidirectional image-sentence retrieval task, supporting up to four languages in a single model. In addition, we show that Machine Translation can be used for data augmentation in multilingual learning, which, combined with MULE, improves mean recall by up to 20.2% on a single language compared to prior work, with the most significant gains seen on languages with relatively few annotations. Our code is publicly available1.

Unsupervised domain adaptation methods traditionally assume that all source categories are present in the target domain. In practice, little may be known about the category overlap between the two domains. While some methods address target settings with either partial or open-set categories, they assume that the particular setting is known a priori. We propose a more universally applicable domain adaptation approach that can handle arbitrary category shift, called Domain Adaptative Neighborhood Clustering via Entropy optimization (DANCE). Our approach combines two novel ideas: First, as we cannot fully rely on source categories to learn features discriminative for the target, we propose a novel neighborhood clustering technique to learn the structure of the target domain in a self-supervised way. Second, we use entropy-based feature alignment and rejection to align target features with the source, or reject them as unknown categories based on their entropy. We show through extensive experiments that DANCE outperforms baselines across open-set, open-partial and partial domain adaptation settings.

Deep models are state-of-the-art for many computer vision tasks including image classification and object detection. However, it has been shown that deep models are vulnerable to adversarial examples. We highlight how one-hot encoding directly contributes to this vulnerability and propose breaking away from this widely-used, but highly-vulnerable mapping. We demonstrate that by leveraging a different output encoding, multi-way encoding, we decorre-late source and target models, making target models more secure. Our approach makes it more difficult for adversaries to find useful gradients for generating adversarial attacks. We present robustness for black-box and white-box attacks on four benchmark datasets: MNIST, CIFAR-10, CIFAR-100, and SVHN. The strength of our approach is also presented in the form of an attack for model watermarking, raising challenges in detecting stolen models.

Contemporary domain adaptation methods are very effective at aligning feature distributions of source and target domains without any target supervision. However, we show that these techniques perform poorly when even a few labeled examples are available in the target domain. To address this semi-supervised domain adaptation (SSDA) setting, we propose a novel Minimax Entropy (MME) approach that adversarially optimizes an adaptive few-shot model. Our base model consists of a feature encoding network, followed by a classification layer that computes the features' similarity to estimated prototypes (representatives of each class). Adaptation is achieved by alternately maximizing the conditional entropy of unlabeled target data with respect to the classifier and minimizing it with respect to the feature encoder. We empirically demonstrate the superiority of our method over many baselines, including conventional feature alignment and few-shot methods, setting a new state of the art for SSDA. Our code is available at http://cs-people.bu.edu/keisaito/research/MME.html.

We propose a novel 3D face recognition algorithm using a deep convolutional neural network (DCNN) and a 3D face expression augmentation technique. The performance of 2D face recognition algorithms has significantly increased by leveraging the representational power of deep neural networks and the use of large-scale labeled training data. In this paper, we show that transfer learning from a CNN trained on 2D face images can effectively work for 3D face recognition by fine-tuning the CNN with an extremely small number of 3D facial scans. We also propose a 3D face expression augmentation technique which synthesizes a number of different facial expressions from a single 3D face scan. Our proposed method shows excellent recognition results on Bosphorus, BU-3DFE, and 3D-TEC datasets without using hand-crafted features. The 3D face identification using our deep features also scales well for large databases.

We propose a 3D face modeling and recognition system using an RGB-D stream in the presence of large pose changes. In the previous work, all facial data points are registered with a reference to improve the accuracy of 3D face model from a low-resolution depth sequence. This registration often fails when applied to non-frontal faces. It causes inaccurate 3D face models and poor performance of matching. We address this problem by pre-aligning each input face (`frontalization') before the registration, which avoids registration failures. For each frame, our method estimates the 3D face pose, assesses the quality of data, segments the facial region, frontalizes it, and performs an accurate registration with the previous 3D model. The 3D-3D recognition system using accurate 3D models from our method outperforms other face recognition systems and shows 100% rank 1 recognition accuracy on a dataset with 30 subjects.

We aim to reconstruct an accurate neutral 3D face model from an RGB-D video in the presence of extreme expression changes. Since each depth frame, taken by a low-cost sensor, is noisy, point clouds from multiple frames can be registered and aggregated to build an accurate 3D model. However, direct aggregation of multiple data produces erroneous results in natural interaction (e.g., talking and showing expressions). We propose to analyze facial expression from an RGB frame and neutralize the corresponding 3D point cloud if needed. We first estimate the person's expression by fitting blendshape coefficients using 2D facial landmarks for each frame and calculate an expression deformity (expression score). With the estimated expression score, we determine whether an input face is neutral or non-neutral. If the face is non-neutral, we proceed to neutralize the expression of the 3D point cloud in that frame. To neutralize the 3D point cloud of a face, we deform our generic 3D face model by applying the estimated blendshape coefficients, find displacement vectors from the deformed generic face to a neutral generic face, and apply the displacement vectors to the input 3D point cloud. After preprocessing frames in a video, we rank frames based on the expression scores and register the ranked frames into a single 3D model. Our system produces a neutral 3D face model in the presence of extreme expression changes even when neutral faces do not exist in the video.