WACV 2021 Papers

Motion estimation is one of the core challenges in computer vision. With traditional dual-frame approaches, occlusions and out-of-view motions are a limiting factor, especially in the context of environmental perception for vehicles due to the large (ego-) motion of objects. Our work proposes a novel data-driven approach for temporal fusion of scene flow estimates in a multi-frame setup to overcome the issue of occlusion. Contrary to most previous methods, we do not rely on a constant motion model, but instead learn a generic temporal relation of motion from data. In a second step, a neural network combines bi-directional scene flow estimates from a common reference frame, yielding a refined estimate and a natural byproduct of occlusion masks. This way, our approach provides a fast multi-frame extension for a variety of scene flow estimators, which outperforms the underlying dual-frame approaches.

Efficient pose estimation finds utility in Augmented Reality (AR) and other computer vision applications such as autonomous navigation and robotics, to name a few. A compact and accurate pose estimation methodology is of paramount importance for on-device inference in such applications. Our proposed solution 3DPoseLite, estimates pose of generic objects by utilizing a compact node embedding representation, unlike computationally expensive multi-view and point-cloud representations. The neural network outputs a 3D pose, taking RGB image and its corresponding graph (obtained by skeletonizing the 3D meshes) as inputs. Our approach utilizes node2vec framework to learn low-dimensional representations for nodes in a graph by optimizing a neighborhood preserving objective. We achieve a space and time reduction by a factor of 11x and 3x respectively, with respect to the state-of-the-art approach, PoseFromShape, on benchmark Pascal3D dataset. We also test the performance of our model on unseen data using Pix3D dataset.

3D human pose and shape estimation plays a vital role in many computer vision applications. There are many deep learning based methods attempting to solve the problem only relying on single-view RGB images for training the network. However, since some public datasets are captured from multi-view cameras system, we propose a novel method to tackle the problem by putting optimization-based multi-view model-fitting into a regression-based learning loop from multi-view images. Firstly, a convolutional neural network (CNN) regresses the pose and shape of a parametric human body model (SMPL) from multi-view images. Then, utilizing the regressed pose and shape as initialization, we propose an improved multi-view optimization method based on the SMPLify method (MV-SMPLify) to fit the SMPL model to the multi-view images simultaneously. Subsequently, the optimized parameters can be adopted to supervise the training of the CNN model. This whole process forms a self-supervising framework which can combine the advantages of the CNN approach and the optimization-based approach through a collaborative process. In addition, the multi-view images can provide more comprehensive supervision for the training. Experiments on public datasets qualitatively and quantitatively demonstrate that our method outperforms previous approaches in a number of ways.

Manipulating facial expressions is a challenging task due to fine-grained shape changes produced by facial muscles and the lack of input-output pairs for supervised learning. Unlike previous methods using Generative Adversarial Networks (GAN), which rely on cycle-consistency loss or sparse geometry (landmarks) loss for expression synthesis, we propose a novel GAN framework to exploit 3D dense (depth and surface normals) information for expression manipulation. However, a large-scale dataset containing RGB images with expression annotations and their corresponding depth maps is not available. To this end, we propose to use an off-the-shelf state-of-the-art 3D reconstruction model to estimate the depth and create a large-scale RGB-Depth dataset after a manual data clean-up process. We utilise this dataset to minimise the novel depth consistency loss via adversarial learning (note we do not have ground truth depth maps for generated face images) and the depth categorical loss of synthetic data on the discriminator. In addition, to improve the generalisation and lower the bias of the depth parameters, we propose to use a novel confidence regulariser on the discriminator side of the framework. We extensively performed both quantitative and qualitative evaluations on two publicly available challenging facial expression benchmarks: AffectNet and RaFD. Our experiments demonstrate that the proposed method outperforms the competitive baseline and existing arts by a large margin.

Colorization involves the synthesis of colors while preserving structural content as well as the semantics of the target image. This is a well-explored problem in 2D with many state-of-the-art solutions. We explore a new challenge in the field of colorization where we aim at colorizing multi-modal 3D medical data using style exemplars. To the best of our knowledge, this work is the first of its kind so we discuss the full pipeline in detail and the challenges that it brings for 3D medical data. The colorization of medical MRI volume also entails modality conversion that highlights the robustness of our approach in handling multi-modal data.