Yikang Li

Jiakang Yuan, Bo Zhang, Xiangchao Yan et al. · deepmind

Unsupervised Domain Adaptation (UDA) technique has been explored in 3D cross-domain tasks recently. Though preliminary progress has been made, the performance gap between the UDA-based 3D model and the supervised one trained with fully annotated target domain is still large. This motivates us to consider selecting partial-yet-important target data and labeling them at a minimum cost, to achieve a good trade-off between high performance and low annotation cost. To this end, we propose a Bi-domain active learning approach, namely Bi3D, to solve the cross-domain 3D object detection task. The Bi3D first develops a domainness-aware source sampling strategy, which identifies target-domain-like samples from the source domain to avoid the model being interfered by irrelevant source data. Then a diversity-based target sampling strategy is developed, which selects the most informative subset of target domain to improve the model adaptability to the target domain using as little annotation budget as possible. Experiments are conducted on typical cross-domain adaptation scenarios including cross-LiDAR-beam, cross-country, and cross-sensor, where Bi3D achieves a promising target-domain detection accuracy (89.63% on KITTI) compared with UDAbased work (84.29%), even surpassing the detector trained on the full set of the labeled target domain (88.98%). Our code is available at: https://github.com/PJLabADG/3DTrans.

Peng Ye, Baopu Li, Yikang Li et al. · deepmind

Neural Architecture Search~(NAS) has attracted increasingly more attention in recent years because of its capability to design deep neural networks automatically. Among them, differential NAS approaches such as DARTS, have gained popularity for the search efficiency. However, they suffer from two main issues, the weak robustness to the performance collapse and the poor generalization ability of the searched architectures. To solve these two problems, a simple-but-efficient regularization method, termed as Beta-Decay, is proposed to regularize the DARTS-based NAS searching process. Specifically, Beta-Decay regularization can impose constraints to keep the value and variance of activated architecture parameters from too large. Furthermore, we provide in-depth theoretical analysis on how it works and why it works. Experimental results on NAS-Bench-201 show that our proposed method can help to stabilize the searching process and makes the searched network more transferable across different datasets. In addition, our search scheme shows an outstanding property of being less dependent on training time and data. Comprehensive experiments on a variety of search spaces and datasets validate the effectiveness of the proposed method.

Runnan Chen, Youquan Liu, Lingdong Kong et al.

Contrastive Language-Image Pre-training (CLIP) achieves promising results in 2D zero-shot and few-shot learning. Despite the impressive performance in 2D, applying CLIP to help the learning in 3D scene understanding has yet to be explored. In this paper, we make the first attempt to investigate how CLIP knowledge benefits 3D scene understanding. We propose CLIP2Scene, a simple yet effective framework that transfers CLIP knowledge from 2D image-text pre-trained models to a 3D point cloud network. We show that the pre-trained 3D network yields impressive performance on various downstream tasks, i.e., annotation-free and fine-tuning with labelled data for semantic segmentation. Specifically, built upon CLIP, we design a Semantic-driven Cross-modal Contrastive Learning framework that pre-trains a 3D network via semantic and spatial-temporal consistency regularization. For the former, we first leverage CLIP's text semantics to select the positive and negative point samples and then employ the contrastive loss to train the 3D network. In terms of the latter, we force the consistency between the temporally coherent point cloud features and their corresponding image features. We conduct experiments on SemanticKITTI, nuScenes, and ScanNet. For the first time, our pre-trained network achieves annotation-free 3D semantic segmentation with 20.8% and 25.08% mIoU on nuScenes and ScanNet, respectively. When fine-tuned with 1% or 100% labelled data, our method significantly outperforms other self-supervised methods, with improvements of 8% and 1% mIoU, respectively. Furthermore, we demonstrate the generalizability for handling cross-domain datasets. Code is publicly available https://github.com/runnanchen/CLIP2Scene.

Jiakang Yuan, Bo Zhang, Xiangchao Yan et al. · deepmind

It is a long-term vision for Autonomous Driving (AD) community that the perception models can learn from a large-scale point cloud dataset, to obtain unified representations that can achieve promising results on different tasks or benchmarks. Previous works mainly focus on the self-supervised pre-training pipeline, meaning that they perform the pre-training and fine-tuning on the same benchmark, which is difficult to attain the performance scalability and cross-dataset application for the pre-training checkpoint. In this paper, for the first time, we are committed to building a large-scale pre-training point-cloud dataset with diverse data distribution, and meanwhile learning generalizable representations from such a diverse pre-training dataset. We formulate the point-cloud pre-training task as a semi-supervised problem, which leverages the few-shot labeled and massive unlabeled point-cloud data to generate the unified backbone representations that can be directly applied to many baseline models and benchmarks, decoupling the AD-related pre-training process and downstream fine-tuning task. During the period of backbone pre-training, by enhancing the scene- and instance-level distribution diversity and exploiting the backbone's ability to learn from unknown instances, we achieve significant performance gains on a series of downstream perception benchmarks including Waymo, nuScenes, and KITTI, under different baseline models like PV-RCNN++, SECOND, CenterPoint.

Pengjin Wei, Guohang Yan, Yikang Li et al.

With the development of neural networks and the increasing popularity of automatic driving, the calibration of the LiDAR and the camera has attracted more and more attention. This calibration task is multi-modal, where the rich color and texture information captured by the camera and the accurate three-dimensional spatial information from the LiDAR is incredibly significant for downstream tasks. Current research interests mainly focus on obtaining accurate calibration results through information fusion. However, they seldom analyze whether the calibrated results are correct or not, which could be of significant importance in real-world applications. For example, in large-scale production, the LiDARs and the cameras of each smart car have to get well-calibrated as the car leaves the production line, while in the rest of the car life period, the poses of the LiDARs and cameras should also get continually supervised to ensure the security. To this end, this paper proposes a self-checking algorithm to judge whether the extrinsic parameters are well-calibrated by introducing a binary classification network based on the fused information from the camera and the LiDAR. Moreover, since there is no such dataset for the task in this work, we further generate a new dataset branch from the KITTI dataset tailored for the task. Our experiments on the proposed dataset branch demonstrate the performance of our method. To the best of our knowledge, this is the first work to address the significance of continually checking the calibrated extrinsic parameters for autonomous driving. The code is open-sourced on the Github website at https://github.com/OpenCalib/LiDAR2camera_self-check.

Jingyang Lin, Hang Hua, Ming Chen et al.

Video summarization aims to distill the most important information from a source video to produce either an abridged clip or a textual narrative. Traditionally, different methods have been proposed depending on whether the output is a video or text, thus ignoring the correlation between the two semantically related tasks of visual summarization and textual summarization. We propose a new joint video and text summarization task. The goal is to generate both a shortened video clip along with the corresponding textual summary from a long video, collectively referred to as a cross-modal summary. The generated shortened video clip and text narratives should be semantically well aligned. To this end, we first build a large-scale human-annotated dataset -- VideoXum (X refers to different modalities). The dataset is reannotated based on ActivityNet. After we filter out the videos that do not meet the length requirements, 14,001 long videos remain in our new dataset. Each video in our reannotated dataset has human-annotated video summaries and the corresponding narrative summaries. We then design a novel end-to-end model -- VTSUM-BILP to address the challenges of our proposed task. Moreover, we propose a new metric called VT-CLIPScore to help evaluate the semantic consistency of cross-modality summary. The proposed model achieves promising performance on this new task and establishes a benchmark for future research.

Zhaoyang Xia, Youquan Liu, Xin Li et al.

Training deep models for semantic scene completion (SSC) is challenging due to the sparse and incomplete input, a large quantity of objects of diverse scales as well as the inherent label noise for moving objects. To address the above-mentioned problems, we propose the following three solutions: 1) Redesigning the completion sub-network. We design a novel completion sub-network, which consists of several Multi-Path Blocks (MPBs) to aggregate multi-scale features and is free from the lossy downsampling operations. 2) Distilling rich knowledge from the multi-frame model. We design a novel knowledge distillation objective, dubbed Dense-to-Sparse Knowledge Distillation (DSKD). It transfers the dense, relation-based semantic knowledge from the multi-frame teacher to the single-frame student, significantly improving the representation learning of the single-frame model. 3) Completion label rectification. We propose a simple yet effective label rectification strategy, which uses off-the-shelf panoptic segmentation labels to remove the traces of dynamic objects in completion labels, greatly improving the performance of deep models especially for those moving objects. Extensive experiments are conducted in two public SSC benchmarks, i.e., SemanticKITTI and SemanticPOSS. Our SCPNet ranks 1st on SemanticKITTI semantic scene completion challenge and surpasses the competitive S3CNet by 7.2 mIoU. SCPNet also outperforms previous completion algorithms on the SemanticPOSS dataset. Besides, our method also achieves competitive results on SemanticKITTI semantic segmentation tasks, showing that knowledge learned in the scene completion is beneficial to the segmentation task.

Bo Zhang, Jiakang Yuan, Botian Shi et al.

Current 3D object detection models follow a single dataset-specific training and testing paradigm, which often faces a serious detection accuracy drop when they are directly deployed in another dataset. In this paper, we study the task of training a unified 3D detector from multiple datasets. We observe that this appears to be a challenging task, which is mainly due to that these datasets present substantial data-level differences and taxonomy-level variations caused by different LiDAR types and data acquisition standards. Inspired by such observation, we present a Uni3D which leverages a simple data-level correction operation and a designed semantic-level coupling-and-recoupling module to alleviate the unavoidable data-level and taxonomy-level differences, respectively. Our method is simple and easily combined with many 3D object detection baselines such as PV-RCNN and Voxel-RCNN, enabling them to effectively learn from multiple off-the-shelf 3D datasets to obtain more discriminative and generalizable representations. Experiments are conducted on many dataset consolidation settings including Waymo-nuScenes, nuScenes-KITTI, Waymo-KITTI, and Waymo-nuScenes-KITTI consolidations. Their results demonstrate that Uni3D exceeds a series of individual detectors trained on a single dataset, with a 1.04x parameter increase over a selected baseline detector. We expect this work will inspire the research of 3D generalization since it will push the limits of perceptual performance.

Shichao Xu, Yikang Li, Jenhao Hsiao et al.

In computer vision, multi-label recognition are important tasks with many real-world applications, but classifying previously unseen labels remains a significant challenge. In this paper, we propose a novel algorithm, Aligned Dual moDality ClaSsifier (ADDS), which includes a Dual-Modal decoder (DM-decoder) with alignment between visual and textual features, for open-vocabulary multi-label classification tasks. Then we design a simple and yet effective method called Pyramid-Forwarding to enhance the performance for inputs with high resolutions. Moreover, the Selective Language Supervision is applied to further enhance the model performance. Extensive experiments conducted on several standard benchmarks, NUS-WIDE, ImageNet-1k, ImageNet-21k, and MS-COCO, demonstrate that our approach significantly outperforms previous methods and provides state-of-the-art performance for open-vocabulary multi-label classification, conventional multi-label classification and an extreme case called single-to-multi label classification where models trained on single-label datasets (ImageNet-1k, ImageNet-21k) are tested on multi-label ones (MS-COCO and NUS-WIDE).

Shihao Shao, Yikang Li, Zhouchen Lin et al.

Irreducible Cartesian tensors (ICTs) play a crucial role in the design of equivariant graph neural networks, as well as in theoretical chemistry and chemical physics. Meanwhile, the design space of available linear operations on tensors that preserve symmetry presents a significant challenge. The ICT decomposition and a basis of this equivariant space are difficult to obtain for high-rank tensors. After decades of research, Bonvicini (2024) has recently achieved an explicit ICT decomposition for $n=5$ with factorial time/space complexity. In this work we, for the first time, obtain decomposition matrices for ICTs up to rank $n=9$ with reduced and affordable complexity, by constructing what we call path matrices. The path matrices are obtained via performing chain-like contractions with Clebsch-Gordan matrices following the parentage scheme. We prove and leverage that the concatenation of path matrices is an orthonormal change-of-basis matrix between the Cartesian tensor product space and the spherical direct sum spaces. Furthermore, we identify a complete orthogonal basis for the equivariant space, rather than a spanning set (Pearce-Crump, 2023b), through this path matrices technique. Our method avoids the RREF algorithm and maintains a fully analytical derivation of each ICT decomposition matrix, thereby significantly improving the algorithm's speed to obtain arbitrary rank orthogonal ICT decomposition matrices and orthogonal equivariant bases. We further extend our result to the arbitrary tensor product and direct sum spaces, enabling free design between different spaces while keeping symmetry. The Python code is available at https://github.com/ShihaoShao-GH/ICT-decomposition-and-equivariant-bases, where the $n=6,\dots,9$ ICT decomposition matrices are obtained in 1s, 3s, 11s, and 4m32s on 28-core Intel(R) Xeon(R) Gold 6330 CPU @ 2.00GHz, respectively.

Xinmiao Lin, Yikang Li, Jenhao Hsiao et al.

The popular VQ-VAE models reconstruct images through learning a discrete codebook but suffer from a significant issue in the rapid quality degradation of image reconstruction as the compression rate rises. One major reason is that a higher compression rate induces more loss of visual signals on the higher frequency spectrum which reflect the details on pixel space. In this paper, a Frequency Complement Module (FCM) architecture is proposed to capture the missing frequency information for enhancing reconstruction quality. The FCM can be easily incorporated into the VQ-VAE structure, and we refer to the new model as Frequency Augmented VAE (FA-VAE). In addition, a Dynamic Spectrum Loss (DSL) is introduced to guide the FCMs to balance between various frequencies dynamically for optimal reconstruction. FA-VAE is further extended to the text-to-image synthesis task, and a Cross-attention Autoregressive Transformer (CAT) is proposed to obtain more precise semantic attributes in texts. Extensive reconstruction experiments with different compression rates are conducted on several benchmark datasets, and the results demonstrate that the proposed FA-VAE is able to restore more faithfully the details compared to SOTA methods. CAT also shows improved generation quality with better image-text semantic alignment.

Tao Ma, Yikang Li

Generating images with conditional descriptions gains increasing interests in recent years. However, existing conditional inputs are suffering from either unstructured forms (captions) or limited information and expensive labeling (scene graphs). For a targeted scene, the core items, objects, are usually definite while their interactions are flexible and hard to clearly define. Thus, we introduce a more rational setting, generating a realistic image from the objects and captions. Under this setting, objects explicitly define the critical roles in the targeted images and captions implicitly describe their rich attributes and connections. Correspondingly, a MOC-GAN is proposed to mix the inputs of two modalities to generate realistic images. It firstly infers the implicit relations between object pairs from the captions to build a hidden-state scene graph. So a multi-layer representation containing objects, relations and captions is constructed, where the scene graph provides the structures of the scene and the caption provides the image-level guidance. Then a cascaded attentive generative network is designed to coarse-to-fine generate phrase patch by paying attention to the most relevant words in the caption. In addition, a phrase-wise DAMSM is proposed to better supervise the fine-grained phrase-patch consistency. On COCO dataset, our method outperforms the state-of-the-art methods on both Inception Score and FID while maintaining high visual quality. Extensive experiments demonstrate the unique features of our proposed method.

Yikang Li, Pulkit Goel, Varsha Kuppur Rajendra et al.

Conditional text generation has been a challenging task that is yet to see human-level performance from state-of-the-art models. In this work, we specifically focus on the Commongen benchmark, wherein the aim is to generate a plausible sentence for a given set of input concepts. Despite advances in other tasks, large pre-trained language models that are fine-tuned on this dataset often produce sentences that are syntactically correct but qualitatively deviate from a human understanding of common sense. Furthermore, generated sequences are unable to fulfill such lexical requirements as matching part-of-speech and full concept coverage. In this paper, we explore how commonsense knowledge graphs can enhance model performance, with respect to commonsense reasoning and lexically-constrained decoding. We propose strategies for enhancing the semantic correctness of the generated text, which we accomplish through: extracting commonsense relations from Conceptnet, injecting these relations into the Unified Language Model (UniLM) through attention mechanisms, and enforcing the aforementioned lexical requirements through output constraints. By performing several ablations, we find that commonsense injection enables the generation of sentences that are more aligned with human understanding, while remaining compliant with lexical requirements.

Yi Zhou, Shuyang Sun, Chao Zhang et al.

By assigning each relationship a single label, current approaches formulate the relationship detection as a classification problem. Under this formulation, predicate categories are treated as completely different classes. However, different from the object labels where different classes have explicit boundaries, predicates usually have overlaps in their semantic meanings. For example, sit\_on and stand\_on have common meanings in vertical relationships but different details of how these two objects are vertically placed. In order to leverage the inherent structures of the predicate categories, we propose to first build the language hierarchy and then utilize the Hierarchy Guided Feature Learning (HGFL) strategy to learn better region features of both the coarse-grained level and the fine-grained level. Besides, we also propose the Hierarchy Guided Module (HGM) to utilize the coarse-grained level to guide the learning of fine-grained level features. Experiments show that the proposed simple yet effective method can improve several state-of-the-art baselines by a large margin (up to $33\%$ relative gain) in terms of Recall@50 on the task of Scene Graph Generation in different datasets.

Yikang Li, Tianshu Yu, Baoxin Li

Recognizing Video events in long, complex videos with multiple sub-activities has received persistent attention recently. This task is more challenging than traditional action recognition with short, relatively homogeneous video clips. In this paper, we investigate the problem of recognizing long and complex events with varying action rhythms, which has not been considered in the literature but is a practical challenge. Our work is inspired in part by how humans identify events with varying rhythms: quickly catching frames contributing most to a specific event. We propose a two-stage \emph{end-to-end} framework, in which the first stage selects the most significant frames while the second stage recognizes the event using the selected frames. Our model needs only \emph{event-level labels} in the training stage, and thus is more practical when the sub-activity labels are missing or difficult to obtain. The results of extensive experiments show that our model can achieve significant improvement in event recognition from long videos while maintaining high accuracy even if the test videos suffer from severe rhythm changes. This demonstrates the potential of our method for real-world video-based applications, where test and training videos can differ drastically in rhythms of sub-activities.

Pak Lun Kevin Ding, Yikang Li, Baoxin Li

The research on hashing techniques for visual data is gaining increased attention in recent years due to the need for compact representations supporting efficient search/retrieval in large-scale databases such as online images. Among many possibilities, Mean Average Precision(mAP) has emerged as the dominant performance metric for hashing-based retrieval. One glaring shortcoming of mAP is its inability in balancing retrieval accuracy and utilization of hash codes: pushing a system to attain higher mAP will inevitably lead to poorer utilization of the hash codes. Poor utilization of the hash codes hinders good retrieval because of increased collision of samples in the hash space. This means that a model giving a higher mAP values does not necessarily do a better job in retrieval. In this paper, we introduce a new metric named Mean Local Group Average Precision (mLGAP) for better evaluation of the performance of hashing-based retrieval. The new metric provides a retrieval performance measure that also reconciles the utilization of hash codes, leading to a more practically meaningful performance metric than conventional ones like mAP. To this end, we start by mathematical analysis of the deficiencies of mAP for hashing-based retrieval. We then propose mLGAP and show why it is more appropriate for hashing-based retrieval. Experiments on image retrieval are used to demonstrate the effectiveness of the proposed metric.

Peng Gao, Pan Lu, Hongsheng Li et al.

In this paper, we propose a novel Question-Guided Hybrid Convolution (QGHC) network for Visual Question Answering (VQA). Most state-of-the-art VQA methods fuse the high-level textual and visual features from the neural network and abandon the visual spatial information when learning multi-modal features.To address these problems, question-guided kernels generated from the input question are designed to convolute with visual features for capturing the textual and visual relationship in the early stage. The question-guided convolution can tightly couple the textual and visual information but also introduce more parameters when learning kernels. We apply the group convolution, which consists of question-independent kernels and question-dependent kernels, to reduce the parameter size and alleviate over-fitting. The hybrid convolution can generate discriminative multi-modal features with fewer parameters. The proposed approach is also complementary to existing bilinear pooling fusion and attention based VQA methods. By integrating with them, our method could further boost the performance. Extensive experiments on public VQA datasets validate the effectiveness of QGHC.

Yikang Li, Pak Lun Kevin Ding, Baoxin Li

In this paper, we introduce "Power Linear Unit" (PoLU) which increases the nonlinearity capacity of a neural network and thus helps improving its performance. PoLU adopts several advantages of previously proposed activation functions. First, the output of PoLU for positive inputs is designed to be identity to avoid the gradient vanishing problem. Second, PoLU has a non-zero output for negative inputs such that the output mean of the units is close to zero, hence reducing the bias shift effect. Thirdly, there is a saturation on the negative part of PoLU, which makes it more noise-robust for negative inputs. Furthermore, we prove that PoLU is able to map more portions of every layer's input to the same space by using the power function and thus increases the number of response regions of the neural network. We use image classification for comparing our proposed activation function with others. In the experiments, MNIST, CIFAR-10, CIFAR-100, Street View House Numbers (SVHN) and ImageNet are used as benchmark datasets. The neural networks we implemented include widely-used ELU-Network, ResNet-50, and VGG16, plus a couple of shallow networks. Experimental results show that our proposed activation function outperforms other state-of-the-art models with most networks.

Yantian Zha, Yikang Li, Sriram Gopalakrishnan et al.

Recent advances in visual activity recognition have raised the possibility of applications such as automated video surveillance. Effective approaches for such problems however require the ability to recognize the plans of agents from video information. Although traditional plan recognition algorithms depend on access to sophisticated planning domain models, one recent promising direction involves learning approximated (or shallow) domain models directly from the observed activity sequences DUP. One limitation is that such approaches expect observed action sequences as inputs. In many cases involving vision/sensing from raw data, there is considerable uncertainty about the specific action at any given time point. The most we can expect in such cases is probabilistic information about the action at that point. The input will then be sequences of such observed action distributions. In this work, we address the problem of constructing an effective data-interface that allows a plan recognition module to directly handle such observation distributions. Such an interface works like a bridge between the low-level perception module, and the high-level plan recognition module. We propose two approaches. The first involves resampling the distribution sequences to single action sequences, from which we could learn an action affinity model based on learned action (word) embeddings for plan recognition. The second is to directly learn action distribution embeddings by our proposed Distr2vec (distribution to vector) model, to construct an affinity model for plan recognition.

Yikang Li, Nan Duan, Bolei Zhou et al.

Recently visual question answering (VQA) and visual question generation (VQG) are two trending topics in the computer vision, which have been explored separately. In this work, we propose an end-to-end unified framework, the Invertible Question Answering Network (iQAN), to leverage the complementary relations between questions and answers in images by jointly training the model on VQA and VQG tasks. Corresponding parameter sharing scheme and regular terms are proposed as constraints to explicitly leverage Q,A's dependencies to guide the training process. After training, iQAN can take either question or answer as input, then output the counterpart. Evaluated on the large-scale visual question answering datasets CLEVR and VQA2, our iQAN improves the VQA accuracy over the baselines. We also show the dual learning framework of iQAN can be generalized to other VQA architectures and consistently improve the results over both the VQA and VQG tasks.

Yikang Li, Wanli Ouyang, Bolei Zhou et al.

Object detection, scene graph generation and region captioning, which are three scene understanding tasks at different semantic levels, are tied together: scene graphs are generated on top of objects detected in an image with their pairwise relationship predicted, while region captioning gives a language description of the objects, their attributes, relations, and other context information. In this work, to leverage the mutual connections across semantic levels, we propose a novel neural network model, termed as Multi-level Scene Description Network (denoted as MSDN), to solve the three vision tasks jointly in an end-to-end manner. Objects, phrases, and caption regions are first aligned with a dynamic graph based on their spatial and semantic connections. Then a feature refining structure is used to pass messages across the three levels of semantic tasks through the graph. We benchmark the learned model on three tasks, and show the joint learning across three tasks with our proposed method can bring mutual improvements over previous models. Particularly, on the scene graph generation task, our proposed method outperforms the state-of-art method with more than 3% margin.

Yikang Li, Wanli Ouyang, Xiaogang Wang et al.

As the intermediate level task connecting image captioning and object detection, visual relationship detection started to catch researchers' attention because of its descriptive power and clear structure. It detects the objects and captures their pair-wise interactions with a subject-predicate-object triplet, e.g. person-ride-horse. In this paper, each visual relationship is considered as a phrase with three components. We formulate the visual relationship detection as three inter-connected recognition problems and propose a Visual Phrase guided Convolutional Neural Network (ViP-CNN) to address them simultaneously. In ViP-CNN, we present a Phrase-guided Message Passing Structure (PMPS) to establish the connection among relationship components and help the model consider the three problems jointly. Corresponding non-maximum suppression method and model training strategy are also proposed. Experimental results show that our ViP-CNN outperforms the state-of-art method both in speed and accuracy. We further pretrain ViP-CNN on our cleansed Visual Genome Relationship dataset, which is found to perform better than the pretraining on the ImageNet for this task.