WACV 2021 Papers

Dimensionality reduction plays an important role in computer vision problems since it reduces computational cost and is often capable of yielding more discriminative data representation. In this context, Partial Least Squares (PLS) has presented notable results in tasks such as image classification and neural network optimization. However, PLS is infeasible on large datasets (e.g., ImageNet) because it requires all the data to be in memory in advance, which is often impractical due to hardware limitations. Additionally, this requirement prevents us from employing PLS on streaming applications where the data are being continuously generated. Motivated by this, we propose a novel incremental PLS, named Covariance-free Incremental Partial Least Squares (CIPLS), which learns a low-dimensional representation of the data using a single sample at a time. In contrast to other state-of-the-art approaches, instead of adopting a partially-discriminative or SGD-based model, we extend Nonlinear Iterative Partial Least Squares (NIPALS) -- the standard algorithm used to compute PLS -- for incremental processing. Among the advantages of this approach are the preservation of discriminative information across all components, the possibility of employing its score matrices for feature selection, and its computational efficiency. We validate CIPLS on face verification and image classification tasks, where it outperforms several other incremental dimensionality reduction methods. In the context of feature selection, CIPLS achieves comparable results when compared to state-of-the-art techniques.

Image extrapolation aims at expanding the narrow field of view of a given image patch. Existing models mainly deal with natural scene images of homogeneous regions and have no control of the content generation process. In this work, we study conditional image extrapolation to synthesize new images guided by the input structured text. The text is represented as a graph to specify the objects and their spatial relation to the unknown regions of the image. Inspired by drawing techniques, we propose a progressive generative model of three stages, i.e., generating a coarse bounding-boxes layout, refining it to a finer segmentation layout, and mapping the layout to a realistic output. Such a multi-stage design is shown to facilitate the training process and generate more controllable results. We validate the effectiveness of the proposed method on the face and human clothing dataset in terms of visual results, quantitative evaluations, and flexible controls.

The majority of descriptor-based methods for geometric processing of non-rigid shape rely on hand-crafted descriptors. Recently, learning-based techniques have been shown effective, achieving state-of-the-art results in a variety of tasks. Yet, even though these methods can in principle work directly on raw data, most methods still rely on hand-crafted descriptors at the input layer. In this work, we wish to challenge this practice and use a neural network to learn descriptors directly from the raw mesh. To this end, we introduce two modules into our neural architecture. The first is a local reference frame (LRF) used to explicitly make the features invariant to rigid transformations. The second is continuous convolution kernels that provide robustness to sampling. We show the efficacy of our proposed network in learning on raw meshes using two cornerstone tasks: shape matching, and human body parts segmentation. Our results show superior results over baseline methods that use hand-crafted descriptors.

With the explosion of digital data in recent years, continuously learning new tasks from a stream of data without forgetting previously acquired knowledge has become increasingly important. In this paper, we propose a new continual learning (CL) setting, namely "continual representation learning", which focuses on learning better representation in a continuous way. We also provide two large-scale multi-step benchmarks for biometric identification, where the visual appearance of different classes are highly relevant. In contrast to requiring the model to recognize more learned classes, we aim to learn feature representation that can be better generalized to not only previously unseen images but also unseen classes/identities. For the new setting, we propose a novel approach that performs the knowledge distillation over a large number of identities by applying the neighbourhood selection and consistency relaxation strategies to improve scalability and flexibility of the continual learning model. We demonstrate that existing CL methods can improve the representation in the new setting, and our method achieves better results than the competitors.

In this paper, we consider the problem of unsupervised domain adaptation in the semantic segmentation. There are two primary issues in this field, i.e., what and how to transfer domain knowledge across two domains. Existing methods mainly focus on adapting domain-invariant features (what to transfer) through adversarial learning (how to transfer). Context dependency is essential for semantic segmentation, however, its transferability is still not well understood. Furthermore, how to transfer contextual information across two domains remains unexplored. Motivated by this, we propose a cross-attention mechanism based on self-attention to capture context dependencies between two domains and adapt transferable context. To achieve this goal, we design two cross-domain attention modules to adapt context dependencies from both spatial and channel views. Specifically, the spatial attention module captures local feature dependencies between each position in the source and target image. The channel attention module models semantic dependencies between each pair of cross-domain channel maps. To adapt context dependencies, we further selectively aggregate the context information from two domains. The superiority of our method over existing state-of-the-art methods is empirically proved on "GTA5 to Cityscapes" and "SYNTHIA to Cityscapes".

Weight Normalization (WN) is an essential building block in deep learning. However, even state-of-the-art WN methods need to be combined with activation normalization methods, such as Batch Normalization (BN), to provide the same classification accuracy as BN. In this paper, we aim to circumvent this issue with a weight normalization approach that can be used on its own to provide a classification accuracy competitive to BN. Our approach mimics three fundamental properties of BN, namely, keeping the norm of the weights constant, setting the mean of the weights to zero, and simulating stochastic perturbations due to batch sampling bias. Unlike most of the existing WN methods that rely on "reparametrization", our method directly optimizes the weights with proper constraints and thus can circumvent its serious drawback, gradient explosion. Moreover, we propose an efficient and easy-to-implement algorithm to solve our constrained optimization problem without sacrificing its benefits. The results of classification experiments on three popular benchmark datasets demonstrate that our method is highly competitive with or even better than the state-of-the-art normalization methods.

Designing neural network architectures is a challenging task and knowing which specific layers of a model must be adapted to improve the performance is almost a mystery. In this paper, we introduce a novel theory and metric to identify layers that decrease the test accuracy of the trained models, this identification is done as early as at the beginning of training. In the worst-case, such a layer could lead to a network that can not be trained at all. More precisely, we identified those layers that worsen the performance because they produce conflicting training bundles as we show in our novel theoretical analysis, complemented by our extensive empirical studies. Based on these findings, a novel algorithm is introduced to remove performance decreasing layers automatically. Architectures found by this algorithm achieve a competitive accuracy when compared against the state-of-the-art architectures. While keeping such high accuracy, our approach drastically reduces memory consumption and inference time for different computer vision tasks.

The application of convolutional neural networks (CNNs) and deep learning to different domains has become increasingly popular in the last several years. In particular, such models have been used in the agriculture domain to identify plant species, identify plant stresses, and estimate crop yields. Although there has been much success in applying these techniques to the agriculture domain, these works contain many shortcomings that are hindering their chance for adoption in practice (e.g., lack of domain knowledge, predicting only specific stress types, etc.). We address issues of previous works for the task of plant stress identification by applying a hierarchical classification approach employing confidence as a means to determine the specificity of a classification. This work is a collaboration between computer science and agricultural engineering experts.

Keypoints of objects reflect their concise abstractions, while the corresponding connection links (CL) build the skeleton by detecting the intrinsic relations between keypoints. Existing approaches are typically computationally-intensive, inapplicable for instances belonging to multiple classes, and/or infeasible to simultaneously encode connection information. To address the aforementioned issues, we propose an end-to-end category-implicit Keypoint and Link Prediction Network (KLPNet), which is the first approach for simultaneous semantic keypoint detection (for multi-class instances) and CL rejuvenation. In our KLPNet, a novel Conditional Link Prediction Graph is proposed for link prediction among keypoints that are contingent on a predefined category. Furthermore, a Cross-stage Keypoint Localization Module (CKLM) is introduced to explore feature aggregation for coarse-to-fine keypoint localization. Comprehensive experiments conducted on three publicly available benchmarks demonstrate that our KLPNet consistently outperforms all other state-of-the-art approaches. Furthermore, the experimental results of CL prediction also show the effectiveness of our KLPNet with respect to occlusion problems.

In this paper, we investigate the problem of retrieving images from a database based on a multi-modal (image-text) query. Specifically, the query text prompts some modification in the query image and the task is to retrieve images with the desired modifications. For instance, a user of an e-commerce platform is interested in buying a dress, which should look similar to her friend's dress, but the dress should be of white color with a ribbon sash. In this case, we would like the algorithm to retrieve some dresses with desired modifications in the query dress. We propose an autoencoder based model, ComposeAE, to learn the composition of image and text query for retrieving images. We adopt a deep metric learning approach and learn a metric that pushes composition of source image and text query closer to the target images. We also propose a rotational symmetry constraint on the optimization problem. Our approach is able to outperform the state-of-the-art method TIRG on three benchmark datasets, namely: MIT-States, Fashion200k and Fashion IQ. In order to ensure fair comparison, we introduce strong baselines by enhancing TIRG method. To ensure reproducibility of the results, we publish our code here: https://github.com/ecom-research/ComposeAE.

We present a compositional embedding framework that infers not just a single class per input image, but a set of classes, in the setting of one-shot learning. Specifically, we propose and evaluate several novel models consisting of (1) an embedding function f trained jointly with a "composition" function g that computes set union operations between the classes encoded in two embedding vectors; and (2) embedding f trained jointly with a "query" function h that computes whether the classes encoded in one embedding subsume the classes encoded in another embedding. In contrast to prior work, these models must both perceive the classes associated with the input examples and encode the relationships between different class label sets, and they are trained using only weak one-shot supervision consisting of the label-set relationships among training examples. Experiments on the OmniGlot, Open Images, and COCO datasets show that the proposed compositional embedding models outperform existing embedding methods. Our compositional embedding models have applications to multi-label object recognition for both one-shot and supervised learning.

Gaze redirection aims at manipulating the gaze of a given face image with respect to a desired direction (i.e., a reference angle) and it can be applied to many real life scenarios, such as video-conferencing or taking group photos. However, previous work on this topic mainly suffers of two limitations: (1) Low-quality image generation and (2) Low redirection precision. In this paper, we propose to alleviate these problems by means of a novel gaze redirection framework which exploits both a numerical and a pictorial direction guidance, jointly with a coarse-to-fine learning strategy. Specifically, the coarse branch learns the spatial transformation which warps input image according to desired gaze. On the other hand, the fine-grained branch consists of a generator network with conditional residual image learning and a multi-task discriminator. This second branch reduces the gap between the previously warped image and the ground-truth image and recovers finer texture details. Moreover, we propose a numerical and pictorial guidance module (NPG) which uses a pictorial gazemap description and numerical angles as an extra guide to further improve the precision of gaze redirection. Extensive experiments on a benchmark dataset show that the proposed method outperforms the state-of-the-art approaches in terms of both image quality and redirection precision.

In this paper, we present a novel method Coarse- and Fine-grained Attention Network (CFANet) for generating high-quality crowd density maps and people count estimation by incorporating attention maps to better focus on the crowd area. We devise a from-coarse-to-fine progressive attention mechanism by integrating Crowd Region Recognizer (CRR) and Density Level Estimator (DLE) branch, which can suppress the influence of irrelevant background and assign attention weights according to the crowd density levels, because generating accurate fine-grained attention maps directly is normally difficult. We also employ a multi-level supervision mechanism to assist the backpropagation of gradient and reduce overfitting. Besides, we propose a Background-aware Structural Loss (BSL) to reduce the false recognition ratio while improving the structural similarity to groundtruth. Extensive experiments on commonly used datasets show that our method can not only outperform previous state-of-the-art methods in terms of count accuracy but also improve the image quality of density maps as well as reduce the false recognition ratio.

There is significant progress in recognizing traditional human activities from videos focusing on highly distinctive actions involving discriminative body movements, body-object and/or human-human interactions. Driver's activities are different since they are executed by the same subject with similar body parts movements, resulting in subtle changes. To address this, we propose a novel framework by exploiting the spatiotemporal attention to model the subtle changes. Our model is named Coarse Temporal Attention Network (CTA-Net), in which coarse temporal branches are introduced in a trainable glimpse network. The goal is to allow the glimpse to capture high-level temporal relationships, such as 'during', 'before' and 'after' by focusing on a specific part of a video. These branches also respect the topology of the temporal dynamics in the video, ensuring that different branches learn meaningful spatial and temporal changes. The model then uses an innovative attention mechanism to generate high-level action specific contextual information for activity recognition by exploring the hidden states of an LSTM. The attention mechanism helps in learning to decide the importance of each hidden state for the recognition task by weighing them when constructing the representation of the video. Our approach is evaluated on four publicly accessible datasets and significantly outperforms the state-of-the-art by a considerable margin with only RGB video as input.

While ground truth depth data remains hard to obtain, self-supervised monocular depth estimation methods enjoy growing attention. Much research in this area aims at improving loss functions or network architectures. Most works, however, do not leverage self-supervision to its full potential. They stick to the standard closed world train-test pipeline, assuming the network parameters to be fixed after the training is finished. Such an assumption does not allow to adapt to new scenes, whereas with self-supervision this becomes possible without extra annotations. In this paper, we propose a novel self-supervised Continuous Monocular Depth Adaptation method (CoMoDA), which adapts the pretrained model on a test video on the fly. As opposed to existing test-time refinement methods that use isolated frame triplets, we opt for continuous adaptation, making use of the previous experience from the same scene. We additionally augment the proposed procedure with the experience from the distant past, preventing the model from overfitting and thus forgetting already learnt information. We demonstrate that our method can be used for both intra- and cross-dataset adaptation. By adapting the model from train to test set of the Eigen split of KITTI, we achieve state-of-the-art depth estimation performance and surpass all existing methods using standard architectures. We also show that our method runs 15 times faster than existing test-time refinement methods. The code is available at https://github.com/Yevkuzn/CoMoDA.

The state of the art in semantic segmentation is steadily increasing in performance, resulting in more precise and reliable segmentations in many different applications. However, progress is limited by the cost of generating labels for training, which sometimes requires hours of manual labor for a single image. Because of this, semi-supervised methods have been applied to this task, with varying degrees of success. A key challenge is that common augmentations used in semi-supervised classification are less effective for semantic segmentation. We propose a novel data augmentation mechanism called ClassMix, which generates augmentations by mixing unlabelled samples, by leveraging on the network's predictions for respecting object boundaries. We evaluate this augmentation technique on two common semi-supervised semantic segmentation benchmarks, showing that it attains state-of-the-art results. Lastly, we also provide extensive ablation studies comparing different design decisions and training regimes.

Few-shot classification aims to generalize basic knowledge to recognize novel categories from a few samples. Recent centroid-based methods achieve promising classification performance with the nearest neighbor rule. However, we consider that those methods intrinsically ignore per-class distribution, as the decision boundaries are biased due to the diversity of intra-class variances. Hence, we propose a class-wise metric scaling (CMS) mechanism, which can be applied to both training and testing stages. Concretely, metric scalars are set as learnable parameters in the training stage, helping to learn a more discriminative and transferable feature representation. As for testing, we construct a convex optimization problem to generate an optimal scalar vector for refining the nearest neighbor decisions. Besides, we also involve a low-ranking bilinear pooling layer for improved representation capacity, which further provides significant performance gains. Extensive experiments are conducted on a series of feature extractor backbones, datasets, and testing modes, which have shown consistent improvements compared to prior SOTA methods, e.g., we achieve accuracies of 66.64% and 83.63% for 5-way 1-shot and 5-shot settings on the mini-ImageNet, respectively. Under the semi-supervised inductive mode, results are further up to 78.34% and 87.53%, respectively.

Object detection models perform well at localizing and classifying objects that they are shown during training. However, due to the difficulty and cost associated with creating and annotating detection datasets, trained models detect a limited number of object types with unknown objects treated as background content. This hinders the adoption of conventional detectors in real-world applications like large-scale object matching, visual grounding, visual relation prediction, obstacle detection (where it is more important to determine the presence and location of objects than to find specific types), etc. We propose class-agnostic object detection as a new problem that focuses on detecting objects irrespective of their object-classes. Specifically, the goal is to predict bounding boxes for all objects in an image but not their object-classes. The predicted boxes can then be consumed by another system to perform application-specific classification, retrieval, etc. We propose training and evaluation protocols for benchmarking class-agnostic detectors to advance future research in this domain. Finally, we propose (1) baseline methods and (2) a new adversarial learning framework for class-agnostic detection that forces the model to exclude class-specific information from features used for predictions. Experimental results show that adversarial learning improves class-agnostic detection efficacy.

Object counting which aims to calculate the number of total instances of the given class is a classic but crucial task that can be applied to many applications. Most of the prior works only focus on counting certain classes of objects such as people, cars, animals, etc. However, in recent years, there are lots of applications that need to get the count of the unseen class of objects such as a mechanical arm commanded to grab the novel object. In this paper, we present an effective object counting network, Class-agnostic Few-shot Object Counting Network (CFOCNet), that supports counting arbitrary classes of object unseen during training stage. Instead of counting a pre-defined class, our model is able to count instances based on input reference images and reduces the huge cost of data collection, training and parameter tuning for each new object class. Our model utilizes not only the similarity between query image and reference images but self attending the query image to learn the self-repeatedness. Using a two-stream Resnet that matches features in different scales, our network can automatically learn to aggregate different scales of the matching scores. We evaluate our method on the subset of the COCO dataset that contains 80 classes of objects and many diverse scenes. In the experiments, our network outperforms other methods including detection and some previous works by a large margin. To the best of our knowledge, we are the first that mainly focuses on few-shot object counting in the class-agnostic manner.