Analytical dexterous grasping synthesis is often driven by grasp quality metrics. However, existing metrics possess many problems, such as being computationally expensive, physically inaccurate, and non-differentiable. Moreover, none of them can facilitate the synthesis of non-force-closure grasps, which account for a significant portion of task-oriented grasping such as lid screwing and button pushing. The main challenge behind all the above drawbacks is the difficulty in modeling the complex Grasp Wrench Space (GWS). In this work, we overcome this challenge by proposing a novel GWS estimator, thus enabling gradient-based task-oriented dexterous grasp synthesis for the first time. Our key contribution is a fast, accurate, and differentiable technique to estimate the GWS boundary with good physical interpretability by parallel sampling and mapping, which does not require iterative optimization. Second, based on our differentiable GWS estimator, we derive a task-oriented energy function to enable gradient-based grasp synthesis and a metric to evaluate non-force-closure grasps. Finally, we improve the previous dexterous grasp synthesis pipeline mainly by a novel technique to make nearest-point calculation differentiable, even on mesh edges and vertices. Extensive experiments are performed to verify the efficiency and effectiveness of our methods. Our GWS estimator can run in several milliseconds on GPUs with minimal memory cost, more than three orders of magnitude faster than the classic discretization-based method. Using this GWS estimator, we synthesize 0.1 million dexterous grasps to show that our pipeline can significantly outperform the SOTA method, even in task-unaware force-closure-grasp synthesis. For task-oriented grasp synthesis, we provide some qualitative results.
The mechanism of connecting multimodal signals through self-attention operation is a key factor in the success of multimodal Transformer networks in remote sensing data fusion tasks. However, traditional approaches assume access to all modalities during both training and inference, which can lead to severe degradation when dealing with modal-incomplete inputs in downstream applications. To address this limitation, our proposed approach introduces a novel model for incomplete multimodal learning in the context of remote sensing data fusion. This approach can be used in both supervised and self-supervised pretraining paradigms and leverages the additional learned fusion tokens in combination with Bi-LSTM attention and masked self-attention mechanisms to collect multimodal signals. The proposed approach employs reconstruction and contrastive loss to facilitate fusion in pre-training while allowing for random modality combinations as inputs in network training. Our approach delivers state-of-the-art performance on two multimodal datasets for tasks such as building instance / semantic segmentation and land-cover mapping tasks when dealing with incomplete inputs during inference.
While unsupervised change detection using contrastive learning has been significantly improved the performance of literature techniques, at present, it only focuses on the bi-temporal change detection scenario. Previous state-of-the-art models for image time-series change detection often use features obtained by learning for clustering or training a model from scratch using pseudo labels tailored to each scene. However, these approaches fail to exploit the spatial-temporal information of image time-series or generalize to unseen scenarios. In this work, we propose a two-stage approach to unsupervised change detection in satellite image time-series using contrastive learning with feature tracking. By deriving pseudo labels from pre-trained models and using feature tracking to propagate them among the image time-series, we improve the consistency of our pseudo labels and address the challenges of seasonal changes in long-term remote sensing image time-series. We adopt the self-training algorithm with ConvLSTM on the obtained pseudo labels, where we first use supervised contrastive loss and contrastive random walks to further improve the feature correspondence in space-time. Then a fully connected layer is fine-tuned on the pre-trained multi-temporal features for generating the final change maps. Through comprehensive experiments on two datasets, we demonstrate consistent improvements in accuracy on fitting and inference scenarios.
This paper presents Volumetric Transformer Pose estimator (VTP), the first 3D volumetric transformer framework for multi-view multi-person 3D human pose estimation. VTP aggregates features from 2D keypoints in all camera views and directly learns the spatial relationships in the 3D voxel space in an end-to-end fashion. The aggregated 3D features are passed through 3D convolutions before being flattened into sequential embeddings and fed into a transformer. A residual structure is designed to further improve the performance. In addition, the sparse Sinkhorn attention is empowered to reduce the memory cost, which is a major bottleneck for volumetric representations, while also achieving excellent performance. The output of the transformer is again concatenated with 3D convolutional features by a residual design. The proposed VTP framework integrates the high performance of the transformer with volumetric representations, which can be used as a good alternative to the convolutional backbones. Experiments on the Shelf, Campus and CMU Panoptic benchmarks show promising results in terms of both Mean Per Joint Position Error (MPJPE) and Percentage of Correctly estimated Parts (PCP). Our code will be available.
Video summarization aims to automatically generate a diverse and concise summary which is useful in large-scale video processing. Most of methods tend to adopt self attention mechanism across video frames, which fails to model the diversity of video frames. To alleviate this problem, we revisit the pairwise similarity measurement in self attention mechanism and find that the existing inner-product affinity leads to discriminative features rather than diversified features. In light of this phenomenon, we propose global diverse attention by using the squared Euclidean distance instead to compute the affinities. Moreover, we model the local contextual information by proposing local contextual attention to remove the redundancy in the video. By combining these two attention mechanism, a video \textbf{SUM}marization model with Diversified Contextual Attention scheme is developed and named as SUM-DCA. Extensive experiments are conducted on benchmark data sets to verify the effectiveness and the superiority of SUM-DCA in terms of F-score and rank-based evaluation without any bells and whistles.
Our planet is viewed by satellites through multiple sensors (e.g., multi-spectral, Lidar and SAR) and at different times. Multi-view observations bring us complementary information than the single one. Alternatively, there are common features shared between different views, such as geometry and semantics. Recently, contrastive learning methods have been proposed for the alignment of multi-view remote sensing images and improving the feature representation of single sensor images by modeling view-invariant factors. However, these methods are based on the pretraining of the predefined tasks or just focus on image-level classification. Moreover, these methods lack research on uncertainty estimation. In this work, a pixel-wise contrastive approach based on an unlabeled multi-view setting is proposed to overcome this limitation. This is achieved by the use of contrastive loss in the feature alignment and uniformity between multi-view images. In this approach, a pseudo-Siamese ResUnet is trained to learn a representation that aims to align features from the shifted positive pairs and uniform the induced distribution of the features on the hypersphere. The learned features of multi-view remote sensing images are evaluated on a liner protocol evaluation and an unsupervised change detection task. We analyze key properties of the approach that make it work, finding that the requirement of shift equivariance ensured the success of the proposed approach and the uncertainty estimation of representations leads to performance improvements. Moreover, the performance of multi-view contrastive learning is affected by the choice of different sensors. Results demonstrate both improvements in efficiency and accuracy over the state-of-the-art multi-view contrastive methods.
The effective combination of the complementary information provided by the huge amount of unlabeled multi-sensor data (e.g., Synthetic Aperture Radar (SAR), optical images) is a critical topic in remote sensing. Recently, contrastive learning methods have reached remarkable success in obtaining meaningful feature representations from multi-view data. However, these methods only focus on the image-level features, which may not satisfy the requirement for dense prediction tasks such as the land-cover mapping. In this work, we propose a new self-supervised approach to SAR-optical data fusion that can learn disentangled pixel-wise feature representations directly by taking advantage of both multi-view contrastive loss and the bootstrap your own latent (BYOL) methods. Two key contributions of the proposed approach are a multi-view contrastive loss to encode the multimodal images and a shift operation to reconstruct learned representations for each pixel by building the local consistency between different augmented views. In the experimental period, we first verified the effectiveness of multi-view contrastive loss and BYOL in self-supervised learning on SAR-optical fusion using an image-level classification task. Then we validated the proposed approach on a land-cover mapping task by training it with unlabeled SAR-optical image pairs. There we used labeled data pairs to evaluate the discriminative capability of learned features in downstream tasks. Results show that the proposed approach extracts features that result in higher accuracy and that reduces the dimension of representations with respect to the image-level contrastive learning method.
The vast amount of unlabeled multi-temporal and multi-sensor remote sensing data acquired by the many Earth Observation satellites present a challenge for change detection. Recently, many generative model-based methods have been proposed for remote sensing image change detection on such unlabeled data. However, the high diversities in the learned features weaken the discrimination of the relevant change indicators in unsupervised change detection tasks. Moreover, these methods lack research on massive archived images. In this work, a self-supervised change detection approach based on an unlabeled multi-view setting is proposed to overcome this limitation. This is achieved by the use of a multi-view contrastive loss and an implicit contrastive strategy in the feature alignment between multi-view images. In this approach, a pseudo-Siamese network is trained to regress the output between its two branches pre-trained in a contrastive way on a large dataset of multi-temporal homogeneous or heterogeneous image patches. Finally, the feature distance between the outputs of the two branches is used to define a change measure, which can be analyzed by thresholding to get the final binary change map. Experiments are carried out on five homogeneous and heterogeneous remote sensing image datasets. The proposed SSL approach is compared with other supervised and unsupervised state-of-the-art change detection methods. Results demonstrate both improvements over state-of-the-art unsupervised methods and that the proposed SSL approach narrows the gap between unsupervised and supervised change detection.
An important topic in the community of remote sensing is how to combine the complementary information provided by the huge amount of unlabeled multi-sensor data, such as Synthetic Aperture Radar (SAR) and optical images. Recently, contrastive learning methods have reached remarkable success in obtaining meaningful feature representations from multi-view data. However, these methods only focus on the image-level features, which may not satisfy the requirement for dense prediction tasks such as land-cover mapping. In this work, we propose a new self-supervised approach for SAR-optical data-fusion that can learn disentangled pixel-wise feature representations directly by taking advantage of both multi-view contrastive learning and the BYOL. The two key contributions were proposed for this approach: multi-view contrastive loss to encode the multi-modal images and the shift operation to reconstruct learned representations for each pixel by building the local consistency between different augmented views. With the aim to validate the effectiveness of the proposed approach, we conduct experiments on the land cover mapping task, where we trained the proposed approach using unlabeled SAR-optical image pairs while labeled data pairs were used for the linear classification and finetuning evaluations. We empirically show that the presented approach outperforms the state-of-the-art methods. In particular, it achieves an improvement on both linear classification and finetuning evaluations and reduces the dimension of representations with respect to the image-level contrastive learning method. Moreover, the proposed method is also validated to bring a sharp improvement on SAR-optical feature fusion than the early fusion fashion for the land-cover mapping task.
There are limited studies on the semantic segmentation of high-resolution Polarimetric Synthetic Aperture Radar (PolSAR) images due to the scarcity of training data and the inference of speckle noises. The Gaofen contest has provided open access of a high-quality PolSAR semantic segmentation dataset. Taking this chance, we propose a Multi-path ResNet (MP-ResNet) architecture for the semantic segmentation of high-resolution PolSAR images. Compared to conventional U-shape encoder-decoder convolutional neural network (CNN) architectures, the MP-ResNet learns semantic context with its parallel multi-scale branches, which greatly enlarges its valid receptive fields and improves the embedding of local discriminative features. In addition, MP-ResNet adopts a multi-level feature fusion design in its decoder to make the best use of the features learned from its different branches. Ablation studies show that the MPResNet has significant advantages over its baseline method (FCN with ResNet34). It also surpasses several classic state-of-the-art methods in terms of overall accuracy (OA), mean F1 and fwIoU, whereas its computational costs are not much increased. This CNN architecture can be used as a baseline method for future studies on the semantic segmentation of PolSAR images. The code is available at: https://github.com/ggsDing/SARSeg.