Radars and cameras belong to the most frequently used sensors for advanced driver assistance systems and automated driving research. However, there has been surprisingly little research on radar-camera fusion with neural networks. One of the reasons is a lack of large-scale automotive datasets with radar and unmasked camera data, with the exception of the nuScenes dataset. Another reason is the difficulty of effectively fusing the sparse radar point cloud on the bird's eye view (BEV) plane with the dense images on the perspective plane. The recent trend of camera-based 3D object detection using BEV features has enabled a new type of fusion, which is better suited for radars. In this work, we present RC-BEVFusion, a modular radar-camera fusion network on the BEV plane. We propose BEVFeatureNet, a novel radar encoder branch, and show that it can be incorporated into several state-of-the-art camera-based architectures. We show significant performance gains of up to 28% increase in the nuScenes detection score, which is an important step in radar-camera fusion research. Without tuning our model for the nuScenes benchmark, we achieve the best result among all published methods in the radar-camera fusion category.
Table detection is the task of classifying and localizing table objects within document images. With the recent development in deep learning methods, we observe remarkable success in table detection. However, a significant amount of labeled data is required to train these models effectively. Many semi-supervised approaches are introduced to mitigate the need for a substantial amount of label data. These approaches use CNN-based detectors that rely on anchor proposals and post-processing stages such as NMS. To tackle these limitations, this paper presents a novel end-to-end semi-supervised table detection method that employs the deformable transformer for detecting table objects. We evaluate our semi-supervised method on PubLayNet, DocBank, ICADR-19 and TableBank datasets, and it achieves superior performance compared to previous methods. It outperforms the fully supervised method (Deformable transformer) by +3.4 points on 10\% labels of TableBank-both dataset and the previous CNN-based semi-supervised approach (Soft Teacher) by +1.8 points on 10\% labels of PubLayNet dataset. We hope this work opens new possibilities towards semi-supervised and unsupervised table detection methods.
In this paper, we introduce a novel network that generates semantic, instance, and part segmentation using a shared encoder and effectively fuses them to achieve panoptic-part segmentation. Unifying these three segmentation problems allows for mutually improved and consistent representation learning. To fuse the predictions of all three heads efficiently, we introduce a parameter-free joint fusion module that dynamically balances the logits and fuses them to create panoptic-part segmentation. Our method is evaluated on the Cityscapes Panoptic Parts (CPP) and Pascal Panoptic Parts (PPP) datasets. For CPP, the PartPQ of our proposed model with joint fusion surpasses the previous state-of-the-art by 1.6 and 4.7 percentage points for all areas and segments with parts, respectively. On PPP, our joint fusion outperforms a model using the previous top-down merging strategy by 3.3 percentage points in PartPQ and 10.5 percentage points in PartPQ for partitionable classes.
Recent works have shown that unstructured text (documents) from online sources can serve as useful auxiliary information for zero-shot image classification. However, these methods require access to a high-quality source like Wikipedia and are limited to a single source of information. Large Language Models (LLM) trained on web-scale text show impressive abilities to repurpose their learned knowledge for a multitude of tasks. In this work, we provide a novel perspective on using an LLM to provide text supervision for a zero-shot image classification model. The LLM is provided with a few text descriptions from different annotators as examples. The LLM is conditioned on these examples to generate multiple text descriptions for each class(referred to as views). Our proposed model, I2MVFormer, learns multi-view semantic embeddings for zero-shot image classification with these class views. We show that each text view of a class provides complementary information allowing a model to learn a highly discriminative class embedding. Moreover, we show that I2MVFormer is better at consuming the multi-view text supervision from LLM compared to baseline models. I2MVFormer establishes a new state-of-the-art on three public benchmark datasets for zero-shot image classification with unsupervised semantic embeddings.
Object permanence is the concept that objects do not suddenly disappear in the physical world. Humans understand this concept at young ages and know that another person is still there, even though it is temporarily occluded. Neural networks currently often struggle with this challenge. Thus, we introduce explicit object permanence into two stage detection approaches drawing inspiration from particle filters. At the core, our detector uses the predictions of previous frames as additional proposals for the current one at inference time. Experiments confirm the feedback loop improving detection performance by a up to 10.3 mAP with little computational overhead. Our approach is suited to extend two-stage detectors for stabilized and reliable detections even under heavy occlusion. Additionally, the ability to apply our method without retraining an existing model promises wide application in real-world tasks.
Despite monocular 3D object detection having recently made a significant leap forward thanks to the use of pre-trained depth estimators for pseudo-LiDAR recovery, such two-stage methods typically suffer from overfitting and are incapable of explicitly encapsulating the geometric relation between depth and object bounding box. To overcome this limitation, we instead propose OPA-3D, a single-stage, end-to-end, Occlusion-Aware Pixel-Wise Aggregation network that to jointly estimate dense scene depth with depth-bounding box residuals and object bounding boxes, allowing a two-stream detection of 3D objects, leading to significantly more robust detections. Thereby, the geometry stream denoted as the Geometry Stream, combines visible depth and depth-bounding box residuals to recover the object bounding box via explicit occlusion-aware optimization. In addition, a bounding box based geometry projection scheme is employed in an effort to enhance distance perception. The second stream, named as the Context Stream, directly regresses 3D object location and size. This novel two-stream representation further enables us to enforce cross-stream consistency terms which aligns the outputs of both streams, improving the overall performance. Extensive experiments on the public benchmark demonstrate that OPA-3D outperforms state-of-the-art methods on the main Car category, whilst keeping a real-time inference speed. We plan to release all codes and trained models soon.
Realistic reconstruction of two hands interacting with objects is a new and challenging problem that is essential for building personalized Virtual and Augmented Reality environments. Graph Convolutional networks (GCNs) allow for the preservation of the topologies of hands poses and shapes by modeling them as a graph. In this work, we propose the THOR-Net which combines the power of GCNs, Transformer, and self-supervision to realistically reconstruct two hands and an object from a single RGB image. Our network comprises two stages; namely the features extraction stage and the reconstruction stage. In the features extraction stage, a Keypoint RCNN is used to extract 2D poses, features maps, heatmaps, and bounding boxes from a monocular RGB image. Thereafter, this 2D information is modeled as two graphs and passed to the two branches of the reconstruction stage. The shape reconstruction branch estimates meshes of two hands and an object using our novel coarse-to-fine GraFormer shape network. The 3D poses of the hands and objects are reconstructed by the other branch using a GraFormer network. Finally, a self-supervised photometric loss is used to directly regress the realistic textured of each vertex in the hands' meshes. Our approach achieves State-of-the-art results in Hand shape estimation on the HO-3D dataset (10.0mm) exceeding ArtiBoost (10.8mm). It also surpasses other methods in hand pose estimation on the challenging two hands and object (H2O) dataset by 5mm on the left-hand pose and 1 mm on the right-hand pose.
In class-incremental semantic segmentation (CISS), deep learning architectures suffer from the critical problems of catastrophic forgetting and semantic background shift. Although recent works focused on these issues, existing classifier initialization methods do not address the background shift problem and assign the same initialization weights to both background and new foreground class classifiers. We propose to address the background shift with a novel classifier initialization method which employs gradient-based attribution to identify the most relevant weights for new classes from the classifier's weights for the previous background and transfers these weights to the new classifier. This warm-start weight initialization provides a general solution applicable to several CISS methods. Furthermore, it accelerates learning of new classes while mitigating forgetting. Our experiments demonstrate significant improvement in mIoU compared to the state-of-the-art CISS methods on the Pascal-VOC 2012, ADE20K and Cityscapes datasets.