Suping Wu's research while affiliated with Ningxia University and other places

Publications (17)

Article
Single-view 3D object reconstruction has always been a long-term challenging task. The objects with complex topologies are hard to accurately be reconstructed, which makes the existing methods suffer from the blurring of the shape boundary between the multiple components in the object. Moreover, most of them cannot balance the learning between glob...
Preprint
In this paper, we present a spatio-temporal tendency reasoning (STR) network for recovering human body pose and shape from videos. Previous approaches have focused on how to extend 3D human datasets and temporal-based learning to promote accuracy and temporal smoothing. Different from them, our STR aims to learn accurate and natural motion sequence...
Chapter
In the unconstrained scene, a large amount of self-occlusion and large poses occurs in the face. These occluded points lead to a huge challenge in face alignment and reconstruction. When performing 3DMM parameter regression, existing face alignment and reconstruction methods only consider reducing the error between the predicted point and the real...
Article
In this paper, we propose a similarity-aware deep adversarial learning (SADAL) approach for facial age estimation. Instead of making full access to the limited training samples which likely leads to bias age prediction, our SADAL aims to seek batches of unobserved hard-negative samples based on existing training samples, which typically reinforces...
Chapter
Plant disease detection plays an important role in agricultural production and ecological protection. However, it is always a challenge to detect the severity of plant diseases in multi-species and multi-disease conditions. Unlike most existing classification methods which are difficult to solve multi-properties detection, we propose a disentangled...
Chapter
Unconstrained face alignment usually undergoes extreme deformations and severe occlusions, which likely gives rise to biased shape prediction. Most existing methods simply exploit shape structure by directly concatenating all landmarks, which leads to losses of facial details in extreme deformation regions. In this paper, we propose a relational-st...
Article
Full-text available
In this paper, we propose an end-to-end reasoning-decision networks (RDN) approach for robust face alignment via policy gradient. Unlike the conventional coarse-to-fine approaches which likely lead to bias prediction due to poor initialization, our approach aims to learn a policy by leveraging raw pixels to reason a subset of shape candidates, sequ...

Citations

... Both of them depend too much on hard-examples mining because most image pairs sampled by the original random sampling strategy satisfy loss-constrained conditions during the late stage of training and make a few contributions to further learning. We do not mine hard-examples in experiments of adding contrastive and triplet loss, which leads to the repeated training process and difficult convergence of the network when optimizing the network by contrastive loss [35] and triplet loss [55]. Therefore, the improvement is slight. ...
... Inspired by the reinforcement learning for face-alignment tasks, in this work we propose medical landmark detection as a Markov decision process [49]. We define L = [L 1 , L 2 , · · · , L I ] ∈ R 2×I as a location vector of I points, where L i denotes for the horizontal and vertical coordinates of the i-th landmark, given a medical image I. ...