Diffusion models have achieved remarkable success in text-to-image generation tasks; however, the role of initial noise has been rarely explored. In this study, we identify specific regions within the initial noise image, termed trigger patches, that play a key role for object generation in the resulting images. Notably, these patches are ``universal'' and can be generalized across various positions, seeds, and prompts. To be specific, extracting these patches from one noise and injecting them into another noise leads to object generation in targeted areas. We identify these patches by analyzing the dispersion of object bounding boxes across generated images, leading to the development of a posterior analysis technique. Furthermore, we create a dataset consisting of Gaussian noises labeled with bounding boxes corresponding to the objects appearing in the generated images and train a detector that identifies these patches from the initial noise. To explain the formation of these patches, we reveal that they are outliers in Gaussian noise, and follow distinct distributions through two-sample tests. Finally, we find the misalignment between prompts and the trigger patch patterns can result in unsuccessful image generations. The study proposes a reject-sampling strategy to obtain optimal noise, aiming to improve prompt adherence and positional diversity in image generation.
The challenge of Out-Of-Distribution (OOD) robustness remains a critical hurdle towards deploying deep vision models. Vision-Language Models (VLMs) have recently achieved groundbreaking results. VLM-based open-vocabulary object detection extends the capabilities of traditional object detection frameworks, enabling the recognition and classification of objects beyond predefined categories. Investigating OOD robustness in recent open-vocabulary object detection is essential to increase the trustworthiness of these models. This study presents a comprehensive robustness evaluation of the zero-shot capabilities of three recent open-vocabulary (OV) foundation object detection models: OWL-ViT, YOLO World, and Grounding DINO. Experiments carried out on the robustness benchmarks COCO-O, COCO-DC, and COCO-C encompassing distribution shifts due to information loss, corruption, adversarial attacks, and geometrical deformation, highlighting the challenges of the model's robustness to foster the research for achieving robustness. Project page: https://prakashchhipa.github.io/projects/ovod_robustness
Human visual reasoning is characterized by an ability to identify abstract patterns from only a small number of examples, and to systematically generalize those patterns to novel inputs. This capacity depends in large part on our ability to represent complex visual inputs in terms of both objects and relations. Recent work in computer vision has introduced models with the capacity to extract object-centric representations, leading to the ability to process multi-object visual inputs, but falling short of the systematic generalization displayed by human reasoning. Other recent models have employed inductive biases for relational abstraction to achieve systematic generalization of learned abstract rules, but have generally assumed the presence of object-focused inputs. Here, we combine these two approaches, introducing Object-Centric Relational Abstraction (OCRA), a model that extracts explicit representations of both objects and abstract relations, and achieves strong systematic generalization in tasks (including a novel dataset, CLEVR-ART, with greater visual complexity) involving complex visual displays.
To ensure safe urban driving for autonomous platforms, it is crucial not only to develop high-performance object detection techniques but also to establish a diverse and representative dataset that captures various urban environments and object characteristics. To address these two issues, we have constructed a multi-class 3D LiDAR dataset reflecting diverse urban environments and object characteristics, and developed a robust 3D semi-supervised object detection (SSOD) based on a multiple teachers framework. This SSOD framework categorizes similar classes and assigns specialized teachers to each category. Through collaborative supervision among these category-specialized teachers, the student network becomes increasingly proficient, leading to a highly effective object detector. We propose a simple yet effective augmentation technique, Pie-based Point Compensating Augmentation (PieAug), to enable the teacher network to generate high-quality pseudo-labels. Extensive experiments on the WOD, KITTI, and our datasets validate the effectiveness of our proposed method and the quality of our dataset. Experimental results demonstrate that our approach consistently outperforms existing state-of-the-art 3D semi-supervised object detection methods across all datasets. We plan to release our multi-class LiDAR dataset and the source code available on our Github repository in the near future.
Advances in NERFs have allowed for 3D scene reconstructions and novel view synthesis. Yet, efficiently editing these representations while retaining photorealism is an emerging challenge. Recent methods face three primary limitations: they're slow for interactive use, lack precision at object boundaries, and struggle to ensure multi-view consistency. We introduce IReNe to address these limitations, enabling swift, near real-time color editing in NeRF. Leveraging a pre-trained NeRF model and a single training image with user-applied color edits, IReNe swiftly adjusts network parameters in seconds. This adjustment allows the model to generate new scene views, accurately representing the color changes from the training image while also controlling object boundaries and view-specific effects. Object boundary control is achieved by integrating a trainable segmentation module into the model. The process gains efficiency by retraining only the weights of the last network layer. We observed that neurons in this layer can be classified into those responsible for view-dependent appearance and those contributing to diffuse appearance. We introduce an automated classification approach to identify these neuron types and exclusively fine-tune the weights of the diffuse neurons. This further accelerates training and ensures consistent color edits across different views. A thorough validation on a new dataset, with edited object colors, shows significant quantitative and qualitative advancements over competitors, accelerating speeds by 5x to 500x.
Open-vocabulary object detection (OVD) requires solid modeling of the region-semantic relationship, which could be learned from massive region-text pairs. However, such data is limited in practice due to significant annotation costs. In this work, we propose RTGen to generate scalable open-vocabulary region-text pairs and demonstrate its capability to boost the performance of open-vocabulary object detection. RTGen includes both text-to-region and region-to-text generation processes on scalable image-caption data. The text-to-region generation is powered by image inpainting, directed by our proposed scene-aware inpainting guider for overall layout harmony. For region-to-text generation, we perform multiple region-level image captioning with various prompts and select the best matching text according to CLIP similarity. To facilitate detection training on region-text pairs, we also introduce a localization-aware region-text contrastive loss that learns object proposals tailored with different localization qualities. Extensive experiments demonstrate that our RTGen can serve as a scalable, semantically rich, and effective source for open-vocabulary object detection and continue to improve the model performance when more data is utilized, delivering superior performance compared to the existing state-of-the-art methods.
This paper introduces YotoR (You Only Transform One Representation), a novel deep learning model for object detection that combines Swin Transformers and YoloR architectures. Transformers, a revolutionary technology in natural language processing, have also significantly impacted computer vision, offering the potential to enhance accuracy and computational efficiency. YotoR combines the robust Swin Transformer backbone with the YoloR neck and head. In our experiments, YotoR models TP5 and BP4 consistently outperform YoloR P6 and Swin Transformers in various evaluations, delivering improved object detection performance and faster inference speeds than Swin Transformer models. These results highlight the potential for further model combinations and improvements in real-time object detection with Transformers. The paper concludes by emphasizing the broader implications of YotoR, including its potential to enhance transformer-based models for image-related tasks.
Current state-of-the-art video object segmentation models have achieved great success using supervised learning with massive labeled training datasets. However, these models are trained using a single source domain and evaluated using videos sampled from the same source domain. When these models are evaluated using videos sampled from a different target domain, their performance degrades significantly due to poor domain generalization, i.e., their inability to learn from multi-domain sources simultaneously using traditional supervised learning. In this paper, We propose a dynamically growing tree of sub-networks (DGT) to learn effectively from multi-domain sources. DGT uses a novel lifelong learning technique that allows the model to continuously and effectively learn from new domains without forgetting the previously learned domains. Hence, the model can generalize to out-of-domain videos. The proposed work is evaluated using single-source in-domain (traditional video object segmentation), multi-source in-domain, and multi-source out-of-domain video object segmentation. The results of DGT show a single source in-domain performance gain of 0.2% and 3.5% on the DAVIS16 and DAVIS17 datasets, respectively. However, when DGT is evaluated using in-domain multi-sources, the results show superior performance compared to state-of-the-art video object segmentation and other lifelong learning techniques with an average performance increase in the F-score of 6.9% with minimal catastrophic forgetting. Finally, in the out-of-domain experiment, the performance of DGT is 2.7% and 4% better than state-of-the-art in 1 and 5-shots, respectively.
We present an object-centric approach to empower robots to learn vision-based manipulation skills from human videos. We investigate the problem of imitating robot manipulation from a single human video in the open-world setting, where a robot must learn to manipulate novel objects from one video demonstration. We introduce ORION, an algorithm that tackles the problem by extracting an object-centric manipulation plan from a single RGB-D video and deriving a policy that conditions on the extracted plan. Our method enables the robot to learn from videos captured by daily mobile devices such as an iPad and generalize the policies to deployment environments with varying visual backgrounds, camera angles, spatial layouts, and novel object instances. We systematically evaluate our method on both short-horizon and long-horizon tasks, demonstrating the efficacy of ORION in learning from a single human video in the open world. Videos can be found in the project website https://ut-austin-rpl.github.io/ORION-release.
Capsule networks (CapsNets) were introduced to address convolutional neural networks limitations, learning object-centric representations that are more robust, pose-aware, and interpretable. They organize neurons into groups called capsules, where each capsule encodes the instantiation parameters of an object or one of its parts. Moreover, a routing algorithm connects capsules in different layers, thereby capturing hierarchical part-whole relationships in the data. This thesis investigates the intriguing aspects of CapsNets and focuses on three key questions to unlock their full potential. First, we explore the effectiveness of the routing algorithm, particularly in small-sized networks. We propose a novel method that anneals the number of routing iterations during training, enhancing performance in architectures with fewer parameters. Secondly, we investigate methods to extract more effective first-layer capsules, also known as primary capsules. By exploiting pruned backbones, we aim to improve computational efficiency by reducing the number of capsules while achieving high generalization. This approach reduces CapsNets memory requirements and computational effort. Third, we explore part-relationship learning in CapsNets. Through extensive research, we demonstrate that capsules with low entropy can extract more concise and discriminative part-whole relationships compared to traditional capsule networks, even with reasonable network sizes. Lastly, we showcase how CapsNets can be utilized in real-world applications, including autonomous localization of unmanned aerial vehicles, quaternion-based rotations prediction in synthetic datasets, and lung nodule segmentation in biomedical imaging. The findings presented in this thesis contribute to a deeper understanding of CapsNets and highlight their potential to address complex computer vision challenges.
Collecting and annotating real-world data for the development of object detection models is a time-consuming and expensive process. In the military domain in particular, data collection can also be dangerous or infeasible. Training models on synthetic data may provide a solution for cases where access to real-world training data is restricted. However, bridging the reality gap between synthetic and real data remains a challenge. Existing methods usually build on top of baseline Convolutional Neural Network (CNN) models that have been shown to perform well when trained on real data, but have limited ability to perform well when trained on synthetic data. For example, some architectures allow for fine-tuning with the expectation of large quantities of training data and are prone to overfitting on synthetic data. Related work usually ignores various best practices from object detection on real data, e.g. by training on synthetic data from a single environment with relatively little variation. In this paper we propose a methodology for improving the performance of a pre-trained object detector when training on synthetic data. Our approach focuses on extracting the salient information from synthetic data without forgetting useful features learned from pre-training on real images. Based on the state of the art, we incorporate data augmentation methods and a Transformer backbone. Besides reaching relatively strong performance without any specialized synthetic data transfer methods, we show that our methods improve the state of the art on synthetic data trained object detection for the RarePlanes and DGTA-VisDrone datasets, and reach near-perfect performance on an in-house vehicle detection dataset.
Object detection forms a key component in Unmanned Aerial Vehicles (UAVs) for completing high-level tasks that depend on the awareness of objects on the ground from an aerial perspective. In that scenario, adversarial patch attacks on an onboard object detector can severely impair the performance of upstream tasks. This paper proposes a novel model-agnostic defense mechanism against the threat of adversarial patch attacks in the context of UAV-based object detection. We formulate adversarial patch defense as an occlusion removal task. The proposed defense method can neutralize adversarial patches located on objects of interest, without exposure to adversarial patches during training. Our lightweight single-stage defense approach allows us to maintain a model-agnostic nature, that once deployed does not require to be updated in response to changes in the object detection pipeline. The evaluations in digital and physical domains show the feasibility of our method for deployment in UAV object detection pipelines, by significantly decreasing the Attack Success Ratio without incurring significant processing costs. As a result, the proposed defense solution can improve the reliability of object detection for UAVs.
Multispectral object detection, utilizing both visible (RGB) and thermal infrared (T) modals, has garnered significant attention for its robust performance across diverse weather and lighting conditions. However, effectively exploiting the complementarity between RGB-T modals while maintaining efficiency remains a critical challenge. In this paper, a very simple Group Shuffled Multi-receptive Attention (GSMA) module is proposed to extract and combine multi-scale RGB and thermal features. Then, the extracted multi-modal features are directly integrated with a multi-level path aggregation neck, which significantly improves the fusion effect and efficiency. Meanwhile, multi-modal object detection often adopts union annotations for both modals. This kind of supervision is not sufficient and unfair, since objects observed in one modal may not be seen in the other modal. To solve this issue, Multi-modal Supervision (MS) is proposed to sufficiently supervise RGB-T object detection. Comprehensive experiments on two challenging benchmarks, KAIST and DroneVehicle, demonstrate the proposed model achieves the state-of-the-art accuracy while maintaining competitive efficiency.
Visual-based perception is the key module for autonomous driving. Among those visual perception tasks, video object detection is a primary yet challenging one because of feature degradation caused by fast motion or multiple poses. Current models usually aggregate features from the neighboring frames to enhance the object representations for the task heads to generate more accurate predictions. Though getting better performance, these methods rely on the information from the future frames and suffer from high computational complexity. Meanwhile, the aggregation process is not reconfigurable during the inference time. These issues make most of the existing models infeasible for online applications. To solve these problems, we introduce a stepwise spatial global-local aggregation network. Our proposed models mainly contain three parts: 1). Multi-stage stepwise network gradually refines the predictions and object representations from the previous stage; 2). Spatial global-local aggregation fuses the local information from the neighboring frames and global semantics from the current frame to eliminate the feature degradation; 3). Dynamic aggregation strategy stops the aggregation process early based on the refinement results to remove redundancy and improve efficiency. Extensive experiments on the ImageNet VID benchmark validate the effectiveness and efficiency of our proposed models.
The Segment Anything Model (SAM) marks a notable milestone in segmentation models, highlighted by its robust zero-shot capabilities and ability to handle diverse prompts. SAM follows a pipeline that separates interactive segmentation into image preprocessing through a large encoder and interactive inference via a lightweight decoder, ensuring efficient real-time performance. However, SAM faces stability issues in challenging samples upon this pipeline. These issues arise from two main factors. Firstly, the image preprocessing disables SAM from dynamically using image-level zoom-in strategies to refocus on the target object during interaction. Secondly, the lightweight decoder struggles to sufficiently integrate interactive information with image embeddings. To address these two limitations, we propose FocSAM with a pipeline redesigned on two pivotal aspects. First, we propose Dynamic Window Multi-head Self-Attention (Dwin-MSA) to dynamically refocus SAM's image embeddings on the target object. Dwin-MSA localizes attention computations around the target object, enhancing object-related embeddings with minimal computational overhead. Second, we propose Pixel-wise Dynamic ReLU (P-DyReLU) to enable sufficient integration of interactive information from a few initial clicks that have significant impacts on the overall segmentation results. Experimentally, FocSAM augments SAM's interactive segmentation performance to match the existing state-of-the-art method in segmentation quality, requiring only about 5.6% of this method's inference time on CPUs.
Despite their significant advancements, Multimodal Large Language Models (MLLMs) often generate factually inaccurate information, referred to as hallucination. In this work, we address object hallucinations in MLLMs, where information is generated about an object not present in the input image. We introduce Data-augmented Phrase-level Alignment (DPA), a novel loss which can be applied to instruction-tuned off-the-shelf MLLMs to mitigate hallucinations, while preserving their general vision-language capabilities. To fine-tune MLLMs with DPA, we first generate a set of `hallucinated' and `correct' response pairs through generative data augmentation by selectively altering the ground-truth information of the correct responses at a phrase level. The DPA loss is then used to train MLLMs to reduce the likelihood of hallucinated phrases compared to the correct ones. Our thorough evaluation on various benchmarks confirms the effectiveness of DPA in mitigating hallucination while retaining the out-of-the-box performance of the MLLMs on general tasks. For instance, MLLMs finetuned with DPA, which we refer to as Hallucination Attenuated Language and Vision Assistant (HALVA), improve F1 by up to 13.4% on hallucination visual question-answering and reduce the hallucination rate by up to 4.2% on image description tasks.
We propose a 3D multi-object tracking (MOT) solution using only 2D detections from monocular cameras, which automatically initiates/terminates tracks as well as resolves track appearance-reappearance and occlusions. Moreover, this approach does not require detector retraining when cameras are reconfigured but only the camera matrices of reconfigured cameras need to be updated. Our approach is based on a Bayesian multi-object formulation that integrates track initiation/termination, re-identification, occlusion handling, and data association into a single Bayes filtering recursion. However, the exact filter that utilizes all these functionalities is numerically intractable due to the exponentially growing number of terms in the (multi-object) filtering density, while existing approximations trade-off some of these functionalities for speed. To this end, we develop a more efficient approximation suitable for online MOT by incorporating object features and kinematics into the measurement model, which improves data association and subsequently reduces the number of terms. Specifically, we exploit the 2D detections and extracted features from multiple cameras to provide a better approximation of the multi-object filtering density to realize the track initiation/termination and re-identification functionalities. Further, incorporating a tractable geometric occlusion model based on 2D projections of 3D objects on the camera planes realizes the occlusion handling functionality of the filter. Evaluation of the proposed solution on challenging datasets demonstrates significant improvements and robustness when camera configurations change on-the-fly, compared to existing multi-view MOT solutions. The source code is publicly available at https://github.com/linh-gist/mv-glmb-ab.
Traditional image-to-3D models often struggle with scenes containing multiple objects due to biases and occlusion complexities. To address this challenge, we present REPARO, a novel approach for compositional 3D asset generation from single images. REPARO employs a two-step process: first, it extracts individual objects from the scene and reconstructs their 3D meshes using off-the-shelf image-to-3D models; then, it optimizes the layout of these meshes through differentiable rendering techniques, ensuring coherent scene composition. By integrating optimal transport-based long-range appearance loss term and high-level semantic loss term in the differentiable rendering, REPARO can effectively recover the layout of 3D assets. The proposed method can significantly enhance object independence, detail accuracy, and overall scene coherence. Extensive evaluation of multi-object scenes demonstrates that our REPARO offers a comprehensive approach to address the complexities of multi-object 3D scene generation from single images.
Visual object tracking, which is primarily based on visible light image sequences, encounters numerous challenges in complicated scenarios, such as low light conditions, high dynamic ranges, and background clutter. To address these challenges, incorporating the advantages of multiple visual modalities is a promising solution for achieving reliable object tracking. However, the existing approaches usually integrate multimodal inputs through adaptive local feature interactions, which cannot leverage the full potential of visual cues, thus resulting in insufficient feature modeling. In this study, we propose a novel multimodal hybrid tracker (MMHT) that utilizes frame-event-based data for reliable single object tracking. The MMHT model employs a hybrid backbone consisting of an artificial neural network (ANN) and a spiking neural network (SNN) to extract dominant features from different visual modalities and then uses a unified encoder to align the features across different domains. Moreover, we propose an enhanced transformer-based module to fuse multimodal features using attention mechanisms. With these methods, the MMHT model can effectively construct a multiscale and multidimensional visual feature space and achieve discriminative feature modeling. Extensive experiments demonstrate that the MMHT model exhibits competitive performance in comparison with that of other state-of-the-art methods. Overall, our results highlight the effectiveness of the MMHT model in terms of addressing the challenges faced in visual object tracking tasks.
In real-life scenarios, humans seek out objects in the 3D world to fulfill their daily needs or intentions. This inspires us to introduce 3D intention grounding, a new task in 3D object detection employing RGB-D, based on human intention, such as "I want something to support my back". Closely related, 3D visual grounding focuses on understanding human reference. To achieve detection based on human intention, it relies on humans to observe the scene, reason out the target that aligns with their intention ("pillow" in this case), and finally provide a reference to the AI system, such as "A pillow on the couch". Instead, 3D intention grounding challenges AI agents to automatically observe, reason and detect the desired target solely based on human intention. To tackle this challenge, we introduce the new Intent3D dataset, consisting of 44,990 intention texts associated with 209 fine-grained classes from 1,042 scenes of the ScanNet dataset. We also establish several baselines based on different language-based 3D object detection models on our benchmark. Finally, we propose IntentNet, our unique approach, designed to tackle this intention-based detection problem. It focuses on three key aspects: intention understanding, reasoning to identify object candidates, and cascaded adaptive learning that leverages the intrinsic priority logic of different losses for multiple objective optimization.