Han Hu

Shuyuan Tu, Qi Dai, Zuxuan Wu et al. · cmu, uw

Contrastive language-image pretraining (CLIP) has demonstrated remarkable success in various image tasks. However, how to extend CLIP with effective temporal modeling is still an open and crucial problem. Existing factorized or joint spatial-temporal modeling trades off between the efficiency and performance. While modeling temporal information within straight through tube is widely adopted in literature, we find that simple frame alignment already provides enough essence without temporal attention. To this end, in this paper, we proposed a novel Implicit Learnable Alignment (ILA) method, which minimizes the temporal modeling effort while achieving incredibly high performance. Specifically, for a frame pair, an interactive point is predicted in each frame, serving as a mutual information rich region. By enhancing the features around the interactive point, two frames are implicitly aligned. The aligned features are then pooled into a single token, which is leveraged in the subsequent spatial self-attention. Our method allows eliminating the costly or insufficient temporal self-attention in video. Extensive experiments on benchmarks demonstrate the superiority and generality of our module. Particularly, the proposed ILA achieves a top-1 accuracy of 88.7% on Kinetics-400 with much fewer FLOPs compared with Swin-L and ViViT-H. Code is released at https://github.com/Francis-Rings/ILA .

Shuyuan Tu, Qi Dai, Zhi-Qi Cheng et al. · cmu, uw

Existing diffusion-based video editing models have made gorgeous advances for editing attributes of a source video over time but struggle to manipulate the motion information while preserving the original protagonist's appearance and background. To address this, we propose MotionEditor, a diffusion model for video motion editing. MotionEditor incorporates a novel content-aware motion adapter into ControlNet to capture temporal motion correspondence. While ControlNet enables direct generation based on skeleton poses, it encounters challenges when modifying the source motion in the inverted noise due to contradictory signals between the noise (source) and the condition (reference). Our adapter complements ControlNet by involving source content to transfer adapted control signals seamlessly. Further, we build up a two-branch architecture (a reconstruction branch and an editing branch) with a high-fidelity attention injection mechanism facilitating branch interaction. This mechanism enables the editing branch to query the key and value from the reconstruction branch in a decoupled manner, making the editing branch retain the original background and protagonist appearance. We also propose a skeleton alignment algorithm to address the discrepancies in pose size and position. Experiments demonstrate the promising motion editing ability of MotionEditor, both qualitatively and quantitatively.

Jia Ning, Chen Li, Zheng Zhang et al.

Unlike language tasks, where the output space is usually limited to a set of tokens, the output space of visual tasks is more complicated, making it difficult to build a unified visual model for various visual tasks. In this paper, we seek to unify the output space of visual tasks, so that we can also build a unified model for visual tasks. To this end, we demonstrate a single unified model that simultaneously handles two typical visual tasks of instance segmentation and depth estimation, which have discrete/fixed-length and continuous/varied-length outputs, respectively. We propose several new techniques that take into account the particularity of visual tasks: 1) Soft token. We employ soft token to represent the task output. Unlike hard tokens in the common VQ-VAE which are assigned one-hot to discrete codebooks/vocabularies, the soft token is assigned softly to the codebook embeddings. Soft token can improve the accuracy of both the next token inference and decoding of the task output; 2) Mask augmentation. Many visual tasks have corruption, undefined or invalid values in label annotations, i.e., occluded area of depth maps. We show that a mask augmentation technique can greatly benefit these tasks. With these new techniques and other designs, we show that the proposed general-purpose task-solver can perform both instance segmentation and depth estimation well. Particularly, we achieve 0.279 RMSE on the specific task of NYUv2 depth estimation, setting a new record on this benchmark. The general-purpose task-solver, dubbed AiT, is available at \url{https://github.com/SwinTransformer/AiT}.

Ding Jia, Yuhui Yuan, Haodi He et al.

One-to-one set matching is a key design for DETR to establish its end-to-end capability, so that object detection does not require a hand-crafted NMS (non-maximum suppression) to remove duplicate detections. This end-to-end signature is important for the versatility of DETR, and it has been generalized to broader vision tasks. However, we note that there are few queries assigned as positive samples and the one-to-one set matching significantly reduces the training efficacy of positive samples. We propose a simple yet effective method based on a hybrid matching scheme that combines the original one-to-one matching branch with an auxiliary one-to-many matching branch during training. Our hybrid strategy has been shown to significantly improve accuracy. In inference, only the original one-to-one match branch is used, thus maintaining the end-to-end merit and the same inference efficiency of DETR. The method is named H-DETR, and it shows that a wide range of representative DETR methods can be consistently improved across a wide range of visual tasks, including DeformableDETR, PETRv2, PETR, and TransTrack, among others. The code is available at: https://github.com/HDETR

Anke Tang, Li Shen, Yong Luo et al.

Deep model training on extensive datasets is increasingly becoming cost-prohibitive, prompting the widespread adoption of deep model fusion techniques to leverage knowledge from pre-existing models. From simple weight averaging to more sophisticated methods like AdaMerging, model fusion effectively improves model performance and accelerates the development of new models. However, potential interference between parameters of individual models and the lack of interpretability in the fusion progress remain significant challenges. Existing methods often try to resolve the parameter interference issue by evaluating attributes of parameters, such as their magnitude or sign, or by parameter pruning. In this study, we begin by examining the fine-tuning of linear layers through the lens of subspace analysis and explicitly define parameter interference as an optimization problem to shed light on this subject. Subsequently, we introduce an innovative approach to model fusion called zero-shot Sparse MIxture of Low-rank Experts (SMILE) construction, which allows for the upscaling of source models into an MoE model without extra data or further training. Our approach relies on the observation that fine-tuning mostly keeps the important parts from the pre-training, but it uses less significant or unused areas to adapt to new tasks. Also, the issue of parameter interference, which is intrinsically intractable in the original parameter space, can be managed by expanding the dimensions. We conduct extensive experiments across diverse scenarios, such as image classification and text generation tasks, using full fine-tuning and LoRA fine-tuning, and we apply our method to large language models (CLIP models, Flan-T5 models, and Mistral-7B models), highlighting the adaptability and scalability of SMILE. Code is available at https://github.com/tanganke/fusion_bench

Libin Wang, Han Hu, Qisen Shang et al.

The lack of façade structures in photogrammetric mesh models renders them inadequate for meeting the demands of intricate applications. Moreover, these mesh models exhibit irregular surfaces with considerable geometric noise and texture quality imperfections, making the restoration of structures challenging. To address these shortcomings, we present StructuredMesh, a novel approach for reconstructing façade structures conforming to the regularity of buildings within photogrammetric mesh models. Our method involves capturing multi-view color and depth images of the building model using a virtual camera and employing a deep learning object detection pipeline to semi-automatically extract the bounding boxes of façade components such as windows, doors, and balconies from the color image. We then utilize the depth image to remap these boxes into 3D space, generating an initial façade layout. Leveraging architectural knowledge, we apply binary integer programming (BIP) to optimize the 3D layout's structure, encompassing the positions, orientations, and sizes of all components. The refined layout subsequently informs façade modeling through instance replacement. We conducted experiments utilizing building mesh models from three distinct datasets, demonstrating the adaptability, robustness, and noise resistance of our proposed methodology. Furthermore, our 3D layout evaluation metrics reveal that the optimized layout enhances precision, recall, and F-score by 6.5%, 4.5%, and 5.5%, respectively, in comparison to the initial layout.

Qing Zhu, Cheng Liao, Han Hu et al.

Accurately and efficiently extracting building footprints from a wide range of remote sensed imagery remains a challenge due to their complex structure, variety of scales and diverse appearances. Existing convolutional neural network (CNN)-based building extraction methods are complained that they cannot detect the tiny buildings because the spatial information of CNN feature maps are lost during repeated pooling operations of the CNN, and the large buildings still have inaccurate segmentation edges. Moreover, features extracted by a CNN are always partial which restricted by the size of the respective field, and large-scale buildings with low texture are always discontinuous and holey when extracted. This paper proposes a novel multi attending path neural network (MAP-Net) for accurately extracting multiscale building footprints and precise boundaries. MAP-Net learns spatial localization-preserved multiscale features through a multi-parallel path in which each stage is gradually generated to extract high-level semantic features with fixed resolution. Then, an attention module adaptively squeezes channel-wise features from each path for optimization, and a pyramid spatial pooling module captures global dependency for refining discontinuous building footprints. Experimental results show that MAP-Net outperforms state-of-the-art (SOTA) algorithms in boundary localization accuracy as well as continuity of large buildings. Specifically, our method achieved 0.68\%, 1.74\%, 1.46\% precision, and 1.50\%, 1.53\%, 0.82\% IoU score improvement without increasing computational complexity compared with the latest HRNetv2 on the Urban 3D, Deep Globe and WHU datasets, respectively. The TensorFlow implementation is available at https://github.com/lehaifeng/MAPNet.

Han Hu, Peyman Yadmellat

This paper presents a novel method to generate spatial constraints for motion planning in dynamic environments. Motion planning methods for autonomous driving and mobile robots typically need to rely on the spatial constraints imposed by a map-based global planner to generate a collision-free trajectory. These methods may fail without an offline map or where the map is invalid due to dynamic changes in the environment such as road obstruction, construction, and traffic congestion. To address this problem, triangulation-based methods can be used to obtain a spatial constraint. However, the existing methods fall short when dealing with dynamic environments and may lead the motion planner to an unrecoverable state. In this paper, we propose a new method to generate a sequence of channels across different triangulation mesh topologies to serve as the spatial constraints. This can be applied to motion planning of autonomous vehicles or robots in cluttered, unstructured environments. The proposed method is evaluated and compared with other triangulation-based methods in synthetic and complex scenarios collected from a real-world autonomous driving dataset. We have shown that the proposed method results in a more stable, long-term plan with a higher task completion rate, faster arrival time, a higher rate of successful plans, and fewer collisions compared to existing methods.

Li Chen, Yulin Ding, Saeid Pirasteh et al.

Predicting a landslide susceptibility map (LSM) is essential for risk recognition and disaster prevention. Despite the successful application of data-driven approaches for LSM prediction, most methods generally apply a single global model to predict the LSM for an entire target region. However, in large-scale areas with significant environmental change, various parts of the region hold different landslide-inducing environments, and therefore, should be predicted with respective models. This study first segmented target scenarios into blocks for individual analysis. Then, the critical problem is that in each block with limited samples, conducting training and testing a model is impossible for a satisfactory LSM prediction, especially in dangerous mountainous areas where landslide surveying is expensive. To solve the problem, we trained an intermediate representation by the meta-learning paradigm, which is superior for capturing information valuable for few-shot adaption from LSM tasks. We hypothesized that there are more general and vital concepts concerning landslide causes and are sensitive to variations in input features. Thus, we can quickly few-shot adapt the models from the intermediate representation for different blocks or even unseen tasks using very few exemplar samples. Experimental results on the two study areas demonstrated the validity of our block-wise analysis in large scenarios and revealed the top few-shot adaption performances of the proposed methods.

Qing Zhu, Shengzhi Huang, Han Hu et al.

Detecting the changes of buildings in urban environments is essential. Existing methods that use only nadir images suffer from severe problems of ambiguous features and occlusions between buildings and other regions. Furthermore, buildings in urban environments vary significantly in scale, which leads to performance issues when using single-scale features. To solve these issues, this paper proposes a fused feature pyramid network, which utilizes both color and depth data for the 3D verification of existing buildings 2D footprints from oblique images. First, the color data of oblique images are enriched with the depth information rendered from 3D mesh models. Second, multiscale features are fused in the feature pyramid network to convolve both the color and depth data. Finally, multi-view information from both the nadir and oblique images is used in a robust voting procedure to label changes in existing buildings. Experimental evaluations using both the ISPRS benchmark datasets and Shenzhen datasets reveal that the proposed method outperforms the ResNet and EfficientNet networks by 5\% and 2\%, respectively, in terms of recall rate and precision. We demonstrate that the proposed method can successfully detect all changed buildings; therefore, only those marked as changed need to be manually checked during the pipeline updating procedure; this significantly reduces the manual quality control requirements. Moreover, ablation studies indicate that using depth data, feature pyramid modules, and multi-view voting strategies can lead to clear and progressive improvements.

Qing Zhu, Qisen Shang, Han Hu et al.

Photogrammetric mesh models obtained from aerial oblique images have been widely used for urban reconstruction. However, the photogrammetric meshes also suffer from severe texture problems, especially on the road areas due to occlusion. This paper proposes a structure-aware completion approach to improve the quality of meshes by removing undesired vehicles on the road seamlessly. Specifically, the discontinuous texture atlas is first integrated to a continuous screen space through rendering by the graphics pipeline; the rendering also records necessary mapping for deintegration to the original texture atlas after editing. Vehicle regions are masked by a standard object detection approach, e.g. Faster RCNN. Then, the masked regions are completed guided by the linear structures and regularities in the road region, which is implemented based on Patch Match. Finally, the completed rendered image is deintegrated to the original texture atlas and the triangles for the vehicles are also flattened for improved meshes. Experimental evaluations and analyses are conducted against three datasets, which are captured with different sensors and ground sample distances. The results reveal that the proposed method can quite realistic meshes after removing the vehicles. The structure-aware completion approach for road regions outperforms popular image completion methods and ablation study further confirms the effectiveness of the linear guidance. It should be noted that the proposed method is also capable to handle tiled mesh models for large-scale scenes. Dataset and code are available at vrlab.org.cn/~hanhu/projects/mesh.

Qing Zhu, Zhendong Wang, Han Hu et al.

Integration of aerial and ground images has been proved as an efficient approach to enhance the surface reconstruction in urban environments. However, as the first step, the feature point matching between aerial and ground images is remarkably difficult, due to the large differences in viewpoint and illumination conditions. Previous studies based on geometry-aware image rectification have alleviated this problem, but the performance and convenience of this strategy is limited by several flaws, e.g. quadratic image pairs, segregated extraction of descriptors and occlusions. To address these problems, we propose a novel approach: leveraging photogrammetric mesh models for aerial-ground image matching. The methods of this proposed approach have linear time complexity with regard to the number of images, can explicitly handle low overlap using multi-view images and can be directly injected into off-the-shelf structure-from-motion (SfM) and multi-view stereo (MVS) solutions. First, aerial and ground images are reconstructed separately and initially co-registered through weak georeferencing data. Second, aerial models are rendered to the initial ground views, in which the color, depth and normal images are obtained. Then, the synthesized color images and the corresponding ground images are matched by comparing the descriptors, filtered by local geometrical information, and then propagated to the aerial views using depth images and patch-based matching. Experimental evaluations using various datasets confirm the superior performance of the proposed methods in aerial-ground image matching. In addition, incorporation of the existing SfM and MVS solutions into these methods enables more complete and accurate models to be directly obtained.

Qing Zhu, Lin Chen, Han Hu et al.

Precision mapping of landslide inventory is crucial for hazard mitigation. Most landslides generally co-exist with other confusing geological features, and the presence of such areas can only be inferred unambiguously at a large scale. In addition, local information is also important for the preservation of object boundaries. Aiming to solve this problem, this paper proposes an effective approach to fuse both local and non-local features to surmount the contextual problem. Built upon the U-Net architecture that is widely adopted in the remote sensing community, we utilize two additional modules. The first one uses dilated convolution and the corresponding atrous spatial pyramid pooling, which enlarged the receptive field without sacrificing spatial resolution or increasing memory usage. The second uses a scale attention mechanism to guide the up-sampling of features from the coarse level by a learned weight map. In implementation, the computational overhead against the original U-Net was only a few convolutional layers. Experimental evaluations revealed that the proposed method outperformed state-of-the-art general-purpose semantic segmentation approaches. Furthermore, ablation studies have shown that the two models afforded extensive enhancements in landslide-recognition performance.