One of the key problems in 3D object detection is to reduce the accuracy gap between methods based on LiDAR sensors and those based on monocular cameras. A recently proposed framework for monocular 3D detection based on Pseudo-Stereo has received considerable attention in the community. However, so far these two problems are discovered in existing practices, including (1) monocular depth estimation and Pseudo-Stereo detector must be trained separately, (2) Difficult to be compatible with different stereo detectors and (3) the overall calculation is large, which affects the reasoning speed. In this work, we propose an end-to-end, efficient pseudo-stereo 3D detection framework by introducing a Single-View Diffusion Model (SVDM) that uses a few iterations to gradually deliver right informative pixels to the left image. SVDM allows the entire pseudo-stereo 3D detection pipeline to be trained end-to-end and can benefit from the training of stereo detectors. Afterwards, we further explore the application of SVDM in depth-free stereo 3D detection, and the final framework is compatible with most stereo detectors. Among multiple benchmarks on the KITTI dataset, we achieve new state-of-the-art performance.
Recent advancements in Large Language Models (LLMs) such as GPT4 have displayed exceptional multi-modal capabilities in following open-ended instructions given images. However, the performance of these models heavily relies on design choices such as network structures, training data, and training strategies, and these choices have not been extensively discussed in the literature, making it difficult to quantify progress in this field. To address this issue, this paper presents a systematic and comprehensive study, quantitatively and qualitatively, on training such models. We implement over 20 variants with controlled settings. Concretely, for network structures, we compare different LLM backbones and model designs. For training data, we investigate the impact of data and sampling strategies. For instructions, we explore the influence of diversified prompts on the instruction-following ability of the trained models. For benchmarks, we contribute the first, to our best knowledge, comprehensive evaluation set including both image and video tasks through crowd-sourcing. Based on our findings, we present Lynx, which performs the most accurate multi-modal understanding while keeping the best multi-modal generation ability compared to existing open-sourced GPT4-style models.
We tackle the problems of latent variables identification and "out-of-support" image generation in representation learning. We show that both are possible for a class of decoders that we call additive, which are reminiscent of decoders used for object-centric representation learning (OCRL) and well suited for images that can be decomposed as a sum of object-specific images. We provide conditions under which exactly solving the reconstruction problem using an additive decoder is guaranteed to identify the blocks of latent variables up to permutation and block-wise invertible transformations. This guarantee relies only on very weak assumptions about the distribution of the latent factors, which might present statistical dependencies and have an almost arbitrarily shaped support. Our result provides a new setting where nonlinear independent component analysis (ICA) is possible and adds to our theoretical understanding of OCRL methods. We also show theoretically that additive decoders can generate novel images by recombining observed factors of variations in novel ways, an ability we refer to as Cartesian-product extrapolation. We show empirically that additivity is crucial for both identifiability and extrapolation on simulated data.
Intrinsic image decomposition (IID) is the task that decomposes a natural image into albedo and shade. While IID is typically solved through supervised learning methods, it is not ideal due to the difficulty in observing ground truth albedo and shade in general scenes. Conversely, unsupervised learning methods are currently underperforming supervised learning methods since there are no criteria for solving the ill-posed problems. Recently, light detection and ranging (LiDAR) is widely used due to its ability to make highly precise distance measurements. Thus, we have focused on the utilization of LiDAR, especially LiDAR intensity, to address this issue. In this paper, we propose unsupervised intrinsic image decomposition with LiDAR intensity (IID-LI). Since the conventional unsupervised learning methods consist of image-to-image transformations, simply inputting LiDAR intensity is not an effective approach. Therefore, we design an intensity consistency loss that computes the error between LiDAR intensity and gray-scaled albedo to provide a criterion for the ill-posed problem. In addition, LiDAR intensity is difficult to handle due to its sparsity and occlusion, hence, a LiDAR intensity densification module is proposed. We verified the estimating quality using our own dataset, which include RGB images, LiDAR intensity and human judged annotations. As a result, we achieved an estimation accuracy that outperforms conventional unsupervised learning methods. Dataset link : (https://github.com/ntthilab-cv/NTT-intrinsic-dataset).
Novel Categories Discovery (NCD) tackles the open-world problem of classifying known and clustering novel categories based on the class semantics using partial class space annotated data. Unlike traditional pseudo-label and retraining, we investigate NCD from the novel data probability matrix perspective. We leverage the connection between NCD novel data sampling with provided novel class Multinoulli (categorical) distribution and hypothesize to implicitly achieve semantic-based novel data clustering by learning their class distribution. We propose novel constraints on first-order (mean) and second-order (covariance) statistics of probability matrix features while applying instance-wise information constraints. In particular, we align the neuron distribution (activation patterns) under a large batch of Monte-Carlo novel data sampling by matching their empirical features mean and covariance with the provided Multinoulli-distribution. Simultaneously, we minimize entropy and enforce prediction consistency for each instance. Our simple approach successfully realizes semantic-based novel data clustering provided the semantic similarity between label-unlabeled classes. We demonstrate the discriminative capacity of our approaches in image and video modalities. Moreover, we perform extensive ablation studies regarding data, networks, and our framework components to provide better insights. Our approach maintains ~94%, ~93%, and ~85%, classification accuracy in labeled data while achieving ~90%, ~84%, and ~72% clustering accuracy for novel categories for Cifar10, UCF101, and MPSC-ARL datasets that matches state-of-the-art approaches without any external clustering.
Ultrasound (US) imaging is widely used for biometric measurement and diagnosis of internal organs due to the advantages of being real-time and radiation-free. However, due to high inter-operator variability, resulting images highly depend on operators' experience. In this work, an intelligent robotic sonographer is proposed to autonomously "explore" target anatomies and navigate a US probe to a relevant 2D plane by learning from expert. The underlying high-level physiological knowledge from experts is inferred by a neural reward function, using a ranked pairwise image comparisons approach in a self-supervised fashion. This process can be referred to as understanding the "language of sonography". Considering the generalization capability to overcome inter-patient variations, mutual information is estimated by a network to explicitly extract the task-related and domain features in latent space. Besides, a Gaussian distribution-based filter is developed to automatically evaluate and take the quality of the expert's demonstrations into account. The robotic localization is carried out in coarse-to-fine mode based on the predicted reward associated to B-mode images. To demonstrate the performance of the proposed approach, representative experiments for the "line" target and "point" target are performed on vascular phantom and two ex-vivo animal organ phantoms (chicken heart and lamb kidney), respectively. The results demonstrated that the proposed advanced framework can robustly work on different kinds of known and unseen phantoms.
Composing simple elements into complex concepts is crucial yet challenging, especially for 3D action generation. Existing methods largely rely on extensive neural language annotations to discern composable latent semantics, a process that is often costly and labor-intensive. In this study, we introduce a novel framework to generate compositional actions without reliance on language auxiliaries. Our approach consists of three main components: Action Coupling, Conditional Action Generation, and Decoupling Refinement. Action Coupling utilizes an energy model to extract the attention masks of each sub-action, subsequently integrating two actions using these attentions to generate pseudo-training examples. Then, we employ a conditional generative model, CVAE, to learn a latent space, facilitating the diverse generation. Finally, we propose Decoupling Refinement, which leverages a self-supervised pre-trained model MAE to ensure semantic consistency between the sub-actions and compositional actions. This refinement process involves rendering generated 3D actions into 2D space, decoupling these images into two sub-segments, using the MAE model to restore the complete image from sub-segments, and constraining the recovered images to match images rendered from raw sub-actions. Due to the lack of existing datasets containing both sub-actions and compositional actions, we created two new datasets, named HumanAct-C and UESTC-C, and present a corresponding evaluation metric. Both qualitative and quantitative assessments are conducted to show our efficacy.
Learning about the three-dimensional world from two-dimensional images is a fundamental problem in computer vision. An ideal neural network architecture for such tasks would leverage the fact that objects can be rotated and translated in three dimensions to make predictions about novel images. However, imposing SO(3)-equivariance on two-dimensional inputs is difficult because the group of three-dimensional rotations does not have a natural action on the two-dimensional plane. Specifically, it is possible that an element of SO(3) will rotate an image out of plane. We show that an algorithm that learns a three-dimensional representation of the world from two dimensional images must satisfy certain geometric consistency properties which we formulate as SO(2)-equivariance constraints. We use the induced and restricted representations of SO(2) on SO(3) to construct and classify architectures which satisfy these geometric consistency constraints. We prove that any architecture which respects said consistency constraints can be realized as an instance of our construction. We show that three previously proposed neural architectures for 3D pose prediction are special cases of our construction. We propose a new algorithm that is a learnable generalization of previously considered methods. We test our architecture on three pose predictions task and achieve SOTA results on both the PASCAL3D+ and SYMSOL pose estimation tasks.
The prevalence of large-scale multimodal datasets presents unique challenges in assessing dataset quality. We propose a two-step method to analyze multimodal datasets, which leverages a small seed of human annotation to map each multimodal instance to the modalities required to process it. Our method sheds light on the importance of different modalities in datasets, as well as the relationship between them. We apply our approach to TVQA, a video question-answering dataset, and discover that most questions can be answered using a single modality, without a substantial bias towards any specific modality. Moreover, we find that more than 70% of the questions are solvable using several different single-modality strategies, e.g., by either looking at the video or listening to the audio, highlighting the limited integration of multiple modalities in TVQA. We leverage our annotation and analyze the MERLOT Reserve, finding that it struggles with image-based questions compared to text and audio, but also with auditory speaker identification. Based on our observations, we introduce a new test set that necessitates multiple modalities, observing a dramatic drop in model performance. Our methodology provides valuable insights into multimodal datasets and highlights the need for the development of more robust models.
Multimedia recommendation aims to fuse the multi-modal information of items for feature enrichment to improve the recommendation performance. However, existing methods typically introduce multi-modal information based on collaborative information to improve the overall recommendation precision, while failing to explore its cold-start recommendation performance. Meanwhile, these above methods are only applicable when such multi-modal data is available. To address this problem, this paper proposes a recommendation framework, named Cross-modal Content Inference and Feature Enrichment Recommendation (CIERec), which exploits the multi-modal information to improve its cold-start recommendation performance. Specifically, CIERec first introduces image annotation as the privileged information to help guide the mapping of unified features from the visual space to the semantic space in the training phase. And then CIERec enriches the content representation with the fusion of collaborative, visual, and cross-modal inferred representations, so as to improve its cold-start recommendation performance. Experimental results on two real-world datasets show that the content representations learned by CIERec are able to achieve superior cold-start recommendation performance over existing visually-aware recommendation algorithms. More importantly, CIERec can consistently achieve significant improvements with different conventional visually-aware backbones, which verifies its universality and effectiveness.