This website requires JavaScript.

TexPose: Neural Texture Learning for Self-Supervised 6D Object Pose Estimation

Hanzhi ChenFabian ManhardtNassir NavabBenjamin Busam
Dec 2022
摘要
In this paper, we introduce neural texture learning for 6D object poseestimation from synthetic data and a few unlabelled real images. Our majorcontribution is a novel learning scheme which removes the drawbacks of previousworks, namely the strong dependency on co-modalities or additional refinement.These have been previously necessary to provide training signals forconvergence. We formulate such a scheme as two sub-optimisation problems ontexture learning and pose learning. We separately learn to predict realistictexture of objects from real image collections and learn pose estimation frompixel-perfect synthetic data. Combining these two capabilities allows then tosynthesise photorealistic novel views to supervise the pose estimator withaccurate geometry. To alleviate pose noise and segmentation imperfectionpresent during the texture learning phase, we propose a surfel-basedadversarial training loss together with texture regularisation from syntheticdata. We demonstrate that the proposed approach significantly outperforms therecent state-of-the-art methods without ground-truth pose annotations anddemonstrates substantial generalisation improvements towards unseen scenes.Remarkably, our scheme improves the adopted pose estimators substantially evenwhen initialised with much inferior performance.
展开全部
图表提取

暂无人提供速读十问回答

论文十问由沈向洋博士提出,鼓励大家带着这十个问题去阅读论文,用有用的信息构建认知模型。写出自己的十问回答,还有机会在当前页面展示哦。

Q1论文试图解决什么问题?
Q2这是否是一个新的问题?
Q3这篇文章要验证一个什么科学假设?
0
被引用
笔记
问答