ARTÍCULO
TITULO

Occlusion-Aware Unsupervised Learning of Monocular Depth, Optical Flow and Camera Pose with Geometric Constraints

Qianru Teng    
Yimin Chen and Chen Huang    

Resumen

We present an occlusion-aware unsupervised neural network for jointly learning three low-level vision tasks from monocular videos: depth, optical flow, and camera motion. The system consists of three different predicting sub-networks simultaneously coupled by combined loss terms and is capable of computing each task independently on test samples. Geometric constraints extracted from scene geometry which have traditionally been used in bundle adjustment or pose-graph optimization are formed as various self-supervisory signals during our end-to-end learning approach. Different from prior works, our image reconstruction loss also takes account of optical flow. Moreover, we impose novel 3D flow consistency constraints over the predictions of all the three tasks. By explicitly modeling occlusion and taking utilization of both 2D and 3D geometry relationships, abundant geometric constraints are formed over estimated outputs, enabling the system to capture both low-level representations and high-level cues to infer thinner scene structures. Empirical evaluation on the KITTI dataset demonstrates the effectiveness and improvement of our approach: (1) monocular depth estimation outperforms state-of-the-art unsupervised methods and is comparable to stereo supervised ones; (2) optical flow prediction ranks top among prior works and even beats supervised and traditional ones especially in non-occluded regions; (3) pose estimation outperforms established SLAM systems under comparable input settings with a reasonable margin.

 Artículos similares

       
 
Ivan Gabriel-Martin, Alvaro Sordo-Ward, Luis Garrote and Juan T. García    
This paper focuses on proposing the minimum number of storms necessary to derive the extreme flood hydrographs accurately through event-based modelling. To do so, we analyzed the results obtained by coupling a continuous stochastic weather generator (the... ver más
Revista: Water

 
Karin Gallandat, David Stack, Gabrielle String and Daniele Lantagne    
Sodium hypochlorite (NaOCl) and sodium dichloroisocyanurate (NaDCC) are commonly used for household water treatment (HWT); chlorine dioxide (ClO2) is a potential new HWT option. We compared the residual maintenance of NaOCl, NaDCC, and ClO2 over 24 hours... ver más
Revista: Water

 
Carla Escapa, Ricardo N. Coimbra, Teresa Neuparth, Tiago Torres, Miguel M. Santos and Marta Otero    
In this work, zebrafish embryo bioassays were performed to assess the efficiency of microalgae in the removal of acetaminophen from water. Chlorella sorokiniana (CS), Chlorella vulgaris (CV) and Scenedesmus obliquus (SO) were the strains used for water t... ver más
Revista: Water

 
Momina, Mohd Rafatullah, Suzylawati Ismail and Anees Ahmad    
Batch desorption experiments of methylene blue (MB) dye from a clay adsorbent coating were carried out to evaluate the maximum desorption conditions. Combination of thermal and chemical regeneration techniques were used for the desorption process. The de... ver más
Revista: Water

 
Rifaat Abdel Wahaab, Ahmed Salah and Thomas Grischek    
To meet the increasing water demand and to provide safe drinking water in Egypt, the Holding Company for Water and Wastewater (HCWW) and its affiliated companies have started a program to develop riverbank filtration (RBF) sites in all Egyptian governora... ver más
Revista: Water