Compared to typical multi-sensor systems, monocular 3D object detection has attracted much attention due to its simple configuration. However, there is still a significant gap between LiDAR-based and monocular-based methods. In this paper, we find that the ill-posed nature of monocular imagery can lead to depth ambiguity. Specifically, objects with different depths can appear with the same bounding boxes and similar visual features in the 2D image. Unfortunately, the network cannot accurately distinguish different depths from such non-discriminative visual features, resulting in unstable depth training. To facilitate depth learning, we propose a simple yet effective plug-and-play module, One Bounding Box Multiple Objects (OBMO). Concretely, we add a set of suitable pseudo labels by shifting the 3D bounding box along the viewing frustum. To constrain the pseudo-3D labels to be reasonable, we carefully design two label scoring strategies to represent their quality. In contrast to the original hard depth labels, such soft pseudo labels with quality scores allow the network to learn a reasonable depth range, boosting training stability and thus improving final performance. Extensive experiments on KITTI and Waymo benchmarks show that our method significantly improves state-of-the-art monocular 3D detectors by a significant margin (The improvements under the moderate setting on KITTI validation set are $\mathbf{1.82\sim 10.91\%}$ mAP in BEV and $\mathbf{1.18\sim 9.36\%}$ mAP in 3D}. Codes have been released at https://github.com/mrsempress/OBMO.

Previous work on action representation learning focused on global representations for short video clips. In contrast, many practical applications, such as video alignment, strongly demand learning the intensive representation of long videos. In this paper, we introduce a new framework of contrastive action representation learning (CARL) to learn frame-wise action representation in a self-supervised or weakly-supervised manner, especially for long videos. Specifically, we introduce a simple but effective video encoder that considers both spatial and temporal context by combining convolution and transformer. Inspired by the recent massive progress in self-supervised learning, we propose a new sequence contrast loss (SCL) applied to two related views obtained by expanding a series of spatio-temporal data in two versions. One is the self-supervised version that optimizes embedding space by minimizing KL-divergence between sequence similarity of two augmented views and prior Gaussian distribution of timestamp distance. The other is the weakly-supervised version that builds more sample pairs among videos using video-level labels by dynamic time wrapping (DTW). Experiments on FineGym, PennAction, and Pouring datasets show that our method outperforms previous state-of-the-art by a large margin for downstream fine-grained action classification and even faster inference. Surprisingly, although without training on paired videos like in previous works, our self-supervised version also shows outstanding performance in video alignment and fine-grained frame retrieval tasks.

Existing neural rendering methods for creating human avatars typically either require dense input signals such as video or multi-view images, or leverage a learned prior from large-scale specific 3D human datasets such that reconstruction can be performed with sparse-view inputs. Most of these methods fail to achieve realistic reconstruction when only a single image is available. To enable the data-efficient creation of realistic animatable 3D humans, we propose ELICIT, a novel method for learning human-specific neural radiance fields from a single image. Inspired by the fact that humans can easily reconstruct the body geometry and infer the full-body clothing from a single image, we leverage two priors in ELICIT: 3D geometry prior and visual semantic prior. Specifically, ELICIT introduces the 3D body shape geometry prior from a skinned vertex-based template model (i.e., SMPL) and implements the visual clothing semantic prior with the CLIP-based pre-trained models. Both priors are used to jointly guide the optimization for creating plausible content in the invisible areas. In order to further improve visual details, we propose a segmentation-based sampling strategy that locally refines different parts of the avatar. Comprehensive evaluations on multiple popular benchmarks, including ZJU-MoCAP, Human3.6M, and DeepFashion, show that ELICIT has outperformed current state-of-the-art avatar creation methods when only a single image is available. Code will be public for reseach purpose at https://elicit3d.github.io .

We have a Christmas gift for Harry Potter fans all over the world. In this paper, we present Harry Potter Dialogue (HPD), a dataset that helps train Harry Potter-like dialogue agents. Such a task is typically viewed as a variant of personalized dialogue agents, but they differ significantly in three respects: 1) Harry lived in a virtual world of wizards, thus, real-world commonsense may not apply to Harry's conversations; 2) Harry's behavior is strongly linked to background information in conversations: the scene, its attributes and its relationship to other speakers; and 3) Such backgrounds are dynamically altered as the storyline goes on. The HPD dataset, as the first dataset to facilitate the study of dialogue agent construction for characters within a story, provides rich contextual information about each dialogue session such as scenes, character attributes, and relations. More importantly, all the background information will change over the course of the story. In addition, HPD could support both dialogue generation and retrieval tasks. We evaluate baselines such as Dialog-GPT and BOB to determine the extent to which they can generate Harry Potter-like responses. The experimental results disappoint us in that although the generated responses are fluent, they still seem out of character for Harry. Besides, we validate the current most robust dialogue agent, ChatGPT, which also can't generate plausible Harry-Potter-like responses in some cases, either. Our results suggest that there is much scope for future research.

尽管具有明显的区分靶向分布样本的能力，但深度神经网络在检测异常分布数据方面的性能差。为了解决此缺陷，最先进的解决方案选择在离群值的辅助数据集上训练深网。这些辅助离群值的各种培训标准是根据启发式直觉提出的。但是，我们发现这些直观设计的离群训练标准可能会损害分布学习，并最终导致劣等的表现。为此，我们确定了分布不兼容的三个原因：矛盾的梯度，错误的可能性和分布变化。基于我们的新理解，我们通过调整深层模型和损耗函数的顶级设计，提出一种新的分布检测方法。我们的方法通过减少对分布特征的概率特征的干扰来实现分布兼容性。在几个基准上，我们的方法不仅可以实现最新的分布检测性能，而且还提高了分布精度。

两阶段探测器在3D对象检测中已广受欢迎。大多数两阶段的3D检测器都使用网格点，体素电网或第二阶段的ROI特征提取的采样关键点。但是，这种方法在处理不均匀分布和稀疏的室外点方面效率低下。本文在三个方面解决了这个问题。 1）动态点聚集。我们建议补丁搜索以快速在本地区域中为每个3D提案搜索点。然后，将最远的体素采样采样用于均匀采样点。特别是，体素尺寸沿距离变化，以适应点的不均匀分布。 2）Ro-Graph Poling。我们在采样点上构建本地图，以通过迭代消息传递更好地模型上下文信息和地雷关系。 3）视觉功能增强。我们引入了一种简单而有效的融合策略，以补偿具有有限语义提示的稀疏激光雷达点。基于这些模块，我们将图形R-CNN构建为第二阶段，可以将其应用于现有的一阶段检测器，以始终如一地提高检测性能。广泛的实验表明，图R-CNN的表现优于最新的3D检测模型，而Kitti和Waymo Open DataSet的差距很大。我们在Kitti Bev汽车检测排行榜上排名第一。代码将在\ url {https://github.com/nightmare-n/graphrcnn}上找到。

视觉变压器（VIT）是卷积神经网络（CNN）的强大替代方案，引起了很多关注。最近的工作表明，VIT也容易受到CNN等对抗性例子的影响。为了建立强大的VIT，一种直观的方法是应用对抗训练，因为它已被证明是完成强大CNN的最有效方法之一。但是，对抗性培训的一个主要局限性是其沉重的计算成本。 VIT所采用的自我注意力的机制是计算强度的操作，其费用随输入贴片的数量四次增加，从而使VIT上的对抗性训练更加耗时。在这项工作中，我们首先全面研究了有关各种视觉变压器的快速对抗训练，并说明了效率和鲁棒性之间的关系。然后，为了加快对VIT的对抗训练，我们提出了一种有效的注意力引导的对抗训练机制。具体而言，依靠自我注意的专长，我们在对抗训练过程中以注意引导策略的掉落策略积极地嵌入了每一层的某些斑块嵌入。纤细的自我发场模块大大加速了对VIT的对抗训练。只有65％的快速对抗训练时间，我们与具有挑战性的成像网基准相匹配。

由于基于相交的联盟（IOU）优化维持最终IOU预测度量和损失的一致性，因此它已被广泛用于单级2D对象检测器的回归和分类分支。最近，几种3D对象检测方法采用了基于IOU的优化，并用3D iou直接替换了2D iou。但是，由于复杂的实施和效率低下的向后操作，3D中的这种直接计算非常昂贵。此外，基于3D IOU的优化是优化的，因为它对旋转很敏感，因此可能导致训练不稳定性和检测性能恶化。在本文中，我们提出了一种新型的旋转旋转iou（RDIOU）方法，该方法可以减轻旋转敏感性问题，并在训练阶段与3D IOU相比产生更有效的优化目标。具体而言，我们的RDIOU通过将旋转变量解耦为独立术语，但保留3D iou的几何形状来简化回归参数的复杂相互作用。通过将RDIOU纳入回归和分类分支，鼓励网络学习更精确的边界框，并同时克服分类和回归之间的错位问题。基准Kitti和Waymo开放数据集的广泛实验验证我们的RDIOU方法可以为单阶段3D对象检测带来实质性改进。

由于其低成本和设置简单性，单眼3D检测引起了社区的广泛关注。它以RGB图像为输入，并预测3D空间中的3D框。最具挑战性的子任务在于实例深度估计。以前的工作通常使用直接估计方法。但是，在本文中，我们指出RGB图像的实例深度是非直觉的。它是由视觉深度线索和实例属性线索结合在一起的，因此很难在网络中直接学习。因此，我们建议将实例深度重新调整为实例视觉表面深度（视觉深度）和实例属性深度（属性深度）的组合。视觉深度与对象的外观和图像上的位置有关。相比之下，属性深度依赖于对象的固有属性，这些属性与图像上的对象仿射转换不变。相应地，我们将3D位置的不确定性分解为视觉深度不确定性和属性深度不确定性。通过结合不同类型的深度和相关的不确定性，我们可以获得最终的实例深度。此外，单眼3D检测中的数据增强通常由于身体性质而受到限制，从而阻碍了性能的提高。根据提出的实例深度分解策略，我们可以缓解此问题。对Kitti进行了评估，我们的方法实现了新的最新结果，并且广泛的消融研究验证了我们方法中每个组件的有效性。这些代码在https://github.com/spengliang/did-m3d上发布。

由于其效率，一声神经架构搜索（NAS）已被广泛用于发现架构。但是，先前的研究表明，由于架构之间的操作参数过度共享（即大共享范围），架构的一声绩效估计可能与他们在独立培训中的表现没有很好的相关性。因此，最近的方法构建了更高参数化的超级链，以降低共享程度。但是这些改进的方法引入了大量额外的参数，因此在培训成本和排名质量之间导致不良的权衡。为了减轻上述问题，我们建议将课程学习应用于共享范围（接近），以有效地训练超级网。具体而言，我们在一开始就以很大的共享范围（简单的课程）训练超网，并逐渐降低了超级网的共享程度（更难的课程）。为了支持这种培训策略，我们设计了一个新颖的超级网（闭合性），该超级网（CLESENET）将参数从操作中解耦，以实现灵活的共享方案和可调节的共享范围。广泛的实验表明，与其他一击的超级网络相比，Close可以在不同的计算预算限制中获得更好的排名质量，并且在与各种搜索策略结合使用时能够发现出色的体系结构。代码可从https://github.com/walkerning/aw_nas获得。