Local image feature descriptors have had a tremendous impact on the development and application of computer vision methods. It is therefore unsurprising that significant efforts are being made for learning-based image point descriptors. However, the advantage of learned methods over handcrafted methods in real applications is subtle and more nuanced than expected. Moreover, handcrafted descriptors such as SIFT and SURF still perform better point localization in Structure-from-Motion (SfM) compared to many learned counterparts. In this paper, we propose a very simple and effective approach to learning local image descriptors by using a hand-crafted detector and descriptor. Specifically, we choose to learn only the descriptors, supported by handcrafted descriptors while discarding the point localization head. We optimize the final descriptor by leveraging the knowledge already present in the handcrafted descriptor. Such an approach of optimization allows us to discard learning knowledge already present in non-differentiable functions such as the hand-crafted descriptors and only learn the residual knowledge in the main network branch. This offers 50X convergence speed compared to the standard baseline architecture of SuperPoint while at inference the combined descriptor provides superior performance over the learned and hand-crafted descriptors. This is done with minor increase in the computations over the baseline learned descriptor. Our approach has potential applications in ensemble learning and learning with non-differentiable functions. We perform experiments in matching, camera localization and Structure-from-Motion in order to showcase the advantages of our approach.
In this paper, we showcase the effectiveness of optimizing monocular camera poses as a continuous function of time. The camera poses are represented using an implicit neural function which maps the given time to the corresponding camera pose. The mapped camera poses are then used for the downstream tasks where joint camera pose optimization is also required. While doing so, the network parameters -- that implicitly represent camera poses -- are optimized. We exploit the proposed method in four diverse experimental settings, namely, (1) NeRF from noisy poses; (2) NeRF from asynchronous Events; (3) Visual Simultaneous Localization and Mapping (vSLAM); and (4) vSLAM with IMUs. In all four settings, the proposed method performs significantly better than the compared baselines and the state-of-the-art methods. Additionally, using the assumption of continuous motion, changes in pose may actually live in a manifold that has lower than 6 degrees of freedom (DOF) is also realized. We call this low DOF motion representation as the \emph{intrinsic motion} and use the approach in vSLAM settings, showing impressive camera tracking performance.
Modeling Neural Radiance Fields for fast-moving deformable objects from visual data alone is a challenging problem. A major issue arises due to the high deformation and low acquisition rates. To address this problem, we propose to use event cameras that offer very fast acquisition of visual change in an asynchronous manner. In this work, we develop a novel method to model the deformable neural radiance fields using RGB and event cameras. The proposed method uses the asynchronous stream of events and calibrated sparse RGB frames. In our setup, the camera pose at the individual events required to integrate them into the radiance fields remains unknown. Our method jointly optimizes these poses and the radiance field. This happens efficiently by leveraging the collection of events at once and actively sampling the events during learning. Experiments conducted on both realistically rendered graphics and real-world datasets demonstrate a significant benefit of the proposed method over the state-of-the-art and the compared baseline. This shows a promising direction for modeling deformable neural radiance fields in real-world dynamic scenes.
The essence of video semantic segmentation (VSS) is how to leverage temporal information for prediction. Previous efforts are mainly devoted to developing new techniques to calculate the cross-frame affinities such as optical flow and attention. Instead, this paper contributes from a different angle by mining relations among cross-frame affinities, upon which better temporal information aggregation could be achieved. We explore relations among affinities in two aspects: single-scale intrinsic correlations and multi-scale relations. Inspired by traditional feature processing, we propose Single-scale Affinity Refinement (SAR) and Multi-scale Affinity Aggregation (MAA). To make it feasible to execute MAA, we propose a Selective Token Masking (STM) strategy to select a subset of consistent reference tokens for different scales when calculating affinities, which also improves the efficiency of our method. At last, the cross-frame affinities strengthened by SAR and MAA are adopted for adaptively aggregating temporal information. Our experiments demonstrate that the proposed method performs favorably against state-of-the-art VSS methods. The code is publicly available at https://github.com/GuoleiSun/VSS-MRCFA
Neural implicit fields have recently been shown to represent 3D shapes accurately, opening up various applications in 3D shape analysis. Up to now, such implicit fields for 3D representation are scalar, encoding the signed distance or binary volume occupancy and more recently the unsigned distance. However, the first two can only represent closed shapes, while the unsigned distance has difficulties in accurate and fast shape inference. In this paper, we propose a Neural Vector Field for shape representation in order to overcome the two aforementioned problems. Mapping each point in space to the direction towards the closest surface, we can represent any type of shape. Similarly the shape mesh can be reconstructed by applying the marching cubes algorithm, with proposed small changes, on top of the inferred vector field. We compare the method on ShapeNet where the proposed new neural implicit field shows superior accuracy in representing both closed and open shapes outperforming previous methods.
Boundaries are among the primary visual cues used by human and computer vision systems. One of the key problems in boundary detection is the label representation, which typically leads to class imbalance and, as a consequence, to thick boundaries that require non-differential post-processing steps to be thinned. In this paper, we re-interpret boundaries as 1-D surfaces and formulate a one-to-one vector transform function that allows for training of boundary prediction completely avoiding the class imbalance issue. Specifically, we define the boundary representation at any point as the unit vector pointing to the closest boundary surface. Our problem formulation leads to the estimation of direction as well as richer contextual information of the boundary, and, if desired, the availability of zero-pixel thin boundaries also at training time. Our method uses no hyper-parameter in the training loss and a fixed stable hyper-parameter at inference. We provide theoretical justification/discussions of the vector transform representation. We evaluate the proposed loss method using a standard architecture and show the excellent performance over other losses and representations on several datasets.
The design of more complex and powerful neural network models has significantly advanced the state-of-the-art in local feature detection and description. These advances can be attributed to deeper networks, improved training methodologies through self-supervision, or the introduction of new building blocks, such as graph neural networks for feature matching. However, in the pursuit of increased performance, efficient architectures that generate lightweight descriptors have received surprisingly little attention. In this paper, we investigate the adaptations neural networks for detection and description require in order to enable their use in embedded platforms. To that end, we investigate and adapt network quantization techniques for use in real-time applications. In addition, we revisit common practices in descriptor quantization and propose the use of a binary descriptor normalization layer, enabling the generation of distinctive length-invariant binary descriptors. ZippyPoint, our efficient network, runs at 47.2 fps on the Apple M1 CPU. This is up to 5x faster than other learned detection and description models, making it the only real-time learned network. ZippyPoint consistently outperforms all other binary detection and descriptor methods in visual localization and homography estimation tasks. Code and trained models will be released upon publication.
Traditional domain adaptation addresses the task of adapting a model to a novel target domain under limited or no additional supervision. While tackling the input domain gap, the standard domain adaptation settings assume no domain change in the output space. In semantic prediction tasks, different datasets are often labeled according to different semantic taxonomies. In many real-world settings, the target domain task requires a different taxonomy than the one imposed by the source domain. We therefore introduce the more general taxonomy adaptive domain adaptation (TADA) problem, allowing for inconsistent taxonomies between the two domains. We further propose an approach that jointly addresses the image-level and label-level domain adaptation. On the label-level, we employ a bilateral mixed sampling strategy to augment the target domain, and a relabelling method to unify and align the label spaces. We address the image-level domain gap by proposing an uncertainty-rectified contrastive learning method, leading to more domain-invariant and class discriminative features. We extensively evaluate the effectiveness of our framework under different TADA settings: open taxonomy, coarse-to-fine taxonomy, and partially-overlapping taxonomy. Our framework outperforms previous state-of-the-art by a large margin, while capable of adapting to new target domain taxonomies.
We tackle the low-efficiency flaw of vision transformer caused by the high computational/space complexity in Multi-Head Self-Attention (MHSA). To this end, we propose the Hierarchical MHSA (H-MHSA), whose representation is computed in a hierarchical manner. Specifically, our H-MHSA first learns feature relationships within small grids by viewing image patches as tokens. Then, small grids are merged into larger ones, within which feature relationship is learned by viewing each small grid at the preceding step as a token. This process is iterated to gradually reduce the number of tokens. The H-MHSA module is readily pluggable into any CNN architectures and amenable to training via backpropagation. We call this new backbone TransCNN, and it essentially inherits the advantages of both transformer and CNN. Experiments demonstrate that TransCNN achieves state-of-the-art accuracy for image recognition. Code and pretrained models are available at https://github.com/yun-liu/TransCNN. This technical report will keep updating by adding more experiments.