Tracking objects in three-dimensional space is critical for autonomous driving. To ensure safety while driving, the tracker must be able to reliably track objects across frames and accurately estimate their states such as velocity and acceleration in the present. Existing works frequently focus on the association task while either neglecting the model’s performance on state estimation or deploying complex heuristics to predict the states. In this paper, we propose STT, a Stateful Tracking model built with Transformers, that can consistently track objects in the scenes while also predicting their states accurately. STT consumes rich appearance, geometry, and motion signals through long term history of detections and is jointly optimized for both data association and state estimation tasks. Since the standard tracking metrics like MOTA and MOTP do not capture the combined performance of the two tasks in the wider spectrum of object states, we extend them with new metrics called S-MOTA and MOTPSthat address this limitation. STT achieves competitive real-time performance on the Waymo Open Dataset.

Foundation models have significantly enhanced 2D task performance, and recent works like Bridge3D have successfully applied these models to improve 3D scene understanding through knowledge distillation, marking considerable advancements. Nonetheless, challenges such as the misalignment between 2D and 3D representations and the persistent long-tail distribution in 3D datasets still restrict the effectiveness of knowledge distillation from 2D to 3D using foundation models. To tackle these issues, we introduce a novel SAM-guided tokenization method that seamlessly aligns 3D transformer structures with region-level knowledge distillation, replacing the traditional KNN-based tokenization techniques. Additionally, we implement a group-balanced re-weighting strategy to effectively address the long-tail problem in knowledge distillation. Furthermore, inspired by the recent success of masked feature prediction, our framework incorporates a two-stage masked token prediction process in which the student model predicts both the global embeddings and token-wise local embeddings derived from the teacher models trained in the first stage. Our methodology has been validated across multiple datasets, including SUN RGB-D, ScanNet, and S3DIS, for tasks like 3D object detection and semantic segmentation. The results demonstrate significant improvements over current state-of-the-art self-supervised methods, establishing new benchmarks in this field.

Foundation models have achieved remarkable results in 2D and language tasks like image segmentation, object detection, and visual-language understanding. However, their potential to enrich 3D scene representation learning is largely untapped due to the existence of the domain gap. In this work, we propose an innovative methodology called Bridge3D to address this gap by pre-training 3D models using features, semantic masks, and captions sourced from foundation models. Specifically, our method employs semantic masks from foundation models to guide the masking and reconstruction process for the masked autoencoder, enabling more focused attention on foreground representations. Moreover, we bridge the 3D-text gap at the scene level using image captioning foundation models, thereby facilitating scene-level knowledge distillation. We further extend this bridging effort by introducing an innovative object-level knowledge distillation method that harnesses highly accurate object-level masks and semantic text data from foundation models. Our methodology significantly surpasses the performance of existing state-of-the-art methods in 3D object detection and semantic segmentation tasks. For instance, on the ScanNet dataset, Bridge3D improves the baseline by a notable margin of 6.3%. Code will be available at: https://github.com/Zhimin-C/Bridge3D

Recent state-of-the-art method FlexMatch firstly demonstrated that correctly estimating learning status is crucial for semi-supervised learning (SSL). However, the estimation method proposed by FlexMatch does not take into account imbalanced data, which is the common case for 3D semi-supervised learning. To address this problem, we practically demonstrate that unlabeled data class-level confidence can represent the learning status in the 3D imbalanced dataset. Based on this finding, we present a novel class-level confidence based 3D SSL method. Firstly, a dynamic thresholding strategy is proposed to utilize more unlabeled data, especially for low learning status classes. Then, a re-sampling strategy is designed to avoid biasing toward high learning status classes, which dynamically changes the sampling probability of each class. To show the effectiveness of our method in 3D SSL tasks, we conduct extensive experiments on 3D SSL classification and detection tasks. Our method significantly outperforms state-of-the-art counterparts for both 3D SSL classification and detection tasks in all datasets.

Semi-supervised video action recognition tends to enable deep neural networks to achieve remarkable performance even with very limited labeled data. However, existing methods are mainly transferred from current image-based methods (e.g., FixMatch). Without specifically utilizing the temporal dynamics and inherent multimodal attributes, their results could be suboptimal. To better leverage the encoded temporal information in videos, we introduce temporal gradient as an additional modality for more attentive feature extraction in this paper. To be specific, our method explicitly distills the fine-grained motion representations from temporal gradient (TG) and imposes consistency across different modalities (i.e., RGB and TG). The performance of semi-supervised action recognition is significantly improved without additional computation or parameters during inference. Our method achieves the state-of-the-art performance on three video action recognition benchmarks (i.e., Kinetics-400, UCF-101, and HMDB-51) under several typical semi-supervised settings (i.e., different ratios of labeled data). Code is made available at https://github.com/lambert-x/video-semisup.

Monocular image-based 3D perception has become an active research area in recent years owing to its applications in autonomous driving. Approaches to monocular 3D perception including detection and tracking, however, often yield inferior performance when compared to LiDAR-based techniques. Through systematic analysis, we identified that per-object depth estimation accuracy is a major factor bounding the performance. Motivated by this observation, we propose a multi-level fusion method that combines different representations (RGB and pseudo-LiDAR) and temporal information across multiple frames for objects (tracklets) to enhance per-object depth estimation. Our proposed fusion method achieves the state-of-the-art performance of per-object depth estimation on the Waymo Open Dataset, the KITTI detection dataset, and the KITTI MOT dataset. We further demonstrate that by simply replacing estimated depth with fusion-enhanced depth, we can achieve significant improvements in monocular 3D perception tasks, including detection and tracking.

Cross-modal retrieval aims to learn discriminative and modal-invariant features for data from different modalities. Unlike the existing methods which usually learn from the features extracted by offline networks, in this paper, we propose an approach to jointly train the components of cross-modal retrieval framework with metadata, and enable the network to find optimal features. The proposed end-to-end framework is updated with three loss functions: 1) a novel cross-modal center loss to eliminate cross-modal discrepancy, 2) cross-entropy loss to maximize inter-class variations, and 3) mean-square-error loss to reduce modality variations. In particular, our proposed cross-modal center loss minimizes the distances of features from objects belonging to the same class across all modalities. Extensive experiments have been conducted on the retrieval tasks across multi-modalities including 2D image, 3D point cloud and mesh data. The proposed framework significantly out-performs the state-of-the-art methods for both cross-modal and in-domain retrieval for 3D objects on the ModelNet10 and ModelNet40 datasets.

We propose a semi-supervised learning approach for video classification, VideoSSL, using convolutional neural networks (CNN). Like other computer vision tasks, existing supervised video classification methods demand a large amount of labeled data to attain good performance. However, annotation of a large dataset is expensive and time consuming. To minimize the dependence on a large annotated dataset, our proposed semi-supervised method trains from a small number of labeled examples and exploits two regulatory signals from unlabeled data. The first signal is the pseudo-labels of unlabeled examples computed from the confidences of the CNN being trained. The other is the normalized probabilities, as predicted by an image classifier CNN, that captures the information about appearances of the interesting objects in the video. We show that, under the supervision of these guiding signals from unlabeled examples, a video classification CNN can achieve impressive performances utilizing a small fraction of annotated examples on three publicly available datasets: UCF101, HMDB51, and Kinetics.

Large-scale labeled data are generally required to train deep neural networks in order to obtain better performance in visual feature learning from images or videos for computer vision applications. To avoid extensive cost of collecting and annotating large-scale datasets, as a subset of unsupervised learning methods, self-supervised learning methods are proposed to learn general image and video features from large-scale unlabeled data without using any human-annotated labels. This paper provides an extensive review of deep learning-based self-supervised general visual feature learning methods from images or videos. First, the motivation, general pipeline, and terminologies of this field are described. Then the common deep neural network architectures that used for self-supervised learning are summarized. Next, the schema and evaluation metrics of self-supervised learning methods are reviewed followed by the commonly used datasets for images, videos, audios, and 3D data, as well as the existing self-supervised visual feature learning methods. Finally, quantitative performance comparisons of the reviewed methods on benchmark datasets are summarized and discussed for both image and video feature learning. At last, this paper is concluded and lists a set of promising future directions for self-supervised visual feature learning.

As part of the development of an educational tool that can help students achieve fluency in American Sign Language (ASL) through independent and interactive practice with immediate feedback, this paper introduces a near real-time system to recognize grammatical errors in continuous signing videos without necessarily identifying the entire sequence of signs. Our system automatically recognizes if a performance of ASL sentences contains grammatical errors made by ASL students. We first recognize the ASL grammatical elements including both manual gestures and nonmanual signals independently from multiple modalities (i.e. hand gestures, facial expressions, and head movements) by 3D-ResNet networks. Then the temporal boundaries of grammatical elements from different modalities are examined to detect ASL grammatical mistakes by using a sliding window-based approach. We have collected a dataset of continuous sign language, ASL-HW-RGBD, covering different aspects of ASL grammars for training and testing. Our system is able to recognize grammatical elements on ASL-HW-RGBD from manual gestures, facial expressions, and head movements and successfully detect 8 ASL grammatical mistakes.

Deep neural network-based semantic segmentation generally requires large-scale cost extensive annotations for training to obtain better performance. To avoid pixel-wise segmentation annotations that are needed for most methods, recently some researchers attempted to use object-level labels (e.g., bounding boxes) or image-level labels (e.g., image categories). In this paper, we propose a novel recursive coarse-to-fine semantic segmentation framework based on only image-level category labels. For each image, an initial coarse mask is first generated by a convolutional neural network-based unsupervised foreground segmentation model and then is enhanced by a graph model. The enhanced coarse mask is fed to a fully convolutional neural network to be recursively refined. Unlike the existing image-level label-based semantic segmentation methods, which require labeling of all categories for images that contain multiple types of objects, our framework only needs one label for each image and can handle images that contain multi-category objects. Only trained on ImageNet, our framework achieves comparable performance on the PASCAL VOC dataset with other image-level label-based state-of-the-art methods of semantic segmentation. Furthermore, our framework can be easily extended to foreground object segmentation task and achieves comparable performance with the state-of-the-art supervised methods on the Internet object dataset.

In this paper, we propose an efficient and effective action recognition framework by combining multiple feature models from dynamic image, optical flow and raw frame, with 3D convolutional neural network (CNN). Dynamic image preserves the long-term temporal information, while optical flow captures short-term temporal information, and raw frame represents the appearance information. Experiments demonstrate that dynamic image provides complementary information to raw frame feature and optical flow feature. Furthermore, with the approximate rank pooling, the computation of dynamic images is about 360 times faster than optical flow, and the dynamic image requires far less memory than optical flow and raw frame.