DRWR: A Differentiable Renderer without Rendering for Unsupervised 3D Structure Learning from Silhouette Images

Zhizhong Han, Chao Chen, Yu-Shen Liu, Matthias Zwicker
Proceedings of the 37th International Conference on Machine Learning, PMLR 119:3994-4005, 2020.

Abstract

Differentiable renderers have been used successfully for unsupervised 3D structure learning from 2D images because they can bridge the gap between 3D and 2D. To optimize 3D shape parameters, current renderers rely on pixel-wise losses between rendered images of 3D reconstructions and ground truth images from corresponding viewpoints. Hence they require interpolation of the recovered 3D structure at each pixel, visibility handling, and optionally evaluating a shading model. In contrast, here we propose a Differentiable Renderer Without Rendering (DRWR) that omits these steps. DRWR only relies on a simple but effective loss that evaluates how well the projections of reconstructed 3D point clouds cover the ground truth object silhouette. Specifically, DRWR employs a smooth silhouette loss to pull the projection of each individual 3D point inside the object silhouette, and a structure-aware repulsion loss to push each pair of projections that fall inside the silhouette far away from each other. Although we omit surface interpolation, visibility handling, and shading, our results demonstrate that DRWR achieves state-of-the-art accuracies under widely used benchmarks, outperforming previous methods both qualitatively and quantitatively. In addition, our training times are significantly lower due to the simplicity of DRWR.

Cite this Paper


BibTeX
@InProceedings{pmlr-v119-han20b, title = {{DRWR}: A Differentiable Renderer without Rendering for Unsupervised 3{D} Structure Learning from Silhouette Images}, author = {Han, Zhizhong and Chen, Chao and Liu, Yu-Shen and Zwicker, Matthias}, booktitle = {Proceedings of the 37th International Conference on Machine Learning}, pages = {3994--4005}, year = {2020}, editor = {III, Hal Daumé and Singh, Aarti}, volume = {119}, series = {Proceedings of Machine Learning Research}, month = {13--18 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v119/han20b/han20b.pdf}, url = {http://proceedings.mlr.press/v119/han20b.html}, abstract = {Differentiable renderers have been used successfully for unsupervised 3D structure learning from 2D images because they can bridge the gap between 3D and 2D. To optimize 3D shape parameters, current renderers rely on pixel-wise losses between rendered images of 3D reconstructions and ground truth images from corresponding viewpoints. Hence they require interpolation of the recovered 3D structure at each pixel, visibility handling, and optionally evaluating a shading model. In contrast, here we propose a Differentiable Renderer Without Rendering (DRWR) that omits these steps. DRWR only relies on a simple but effective loss that evaluates how well the projections of reconstructed 3D point clouds cover the ground truth object silhouette. Specifically, DRWR employs a smooth silhouette loss to pull the projection of each individual 3D point inside the object silhouette, and a structure-aware repulsion loss to push each pair of projections that fall inside the silhouette far away from each other. Although we omit surface interpolation, visibility handling, and shading, our results demonstrate that DRWR achieves state-of-the-art accuracies under widely used benchmarks, outperforming previous methods both qualitatively and quantitatively. In addition, our training times are significantly lower due to the simplicity of DRWR.} }
Endnote
%0 Conference Paper %T DRWR: A Differentiable Renderer without Rendering for Unsupervised 3D Structure Learning from Silhouette Images %A Zhizhong Han %A Chao Chen %A Yu-Shen Liu %A Matthias Zwicker %B Proceedings of the 37th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2020 %E Hal Daumé III %E Aarti Singh %F pmlr-v119-han20b %I PMLR %P 3994--4005 %U http://proceedings.mlr.press/v119/han20b.html %V 119 %X Differentiable renderers have been used successfully for unsupervised 3D structure learning from 2D images because they can bridge the gap between 3D and 2D. To optimize 3D shape parameters, current renderers rely on pixel-wise losses between rendered images of 3D reconstructions and ground truth images from corresponding viewpoints. Hence they require interpolation of the recovered 3D structure at each pixel, visibility handling, and optionally evaluating a shading model. In contrast, here we propose a Differentiable Renderer Without Rendering (DRWR) that omits these steps. DRWR only relies on a simple but effective loss that evaluates how well the projections of reconstructed 3D point clouds cover the ground truth object silhouette. Specifically, DRWR employs a smooth silhouette loss to pull the projection of each individual 3D point inside the object silhouette, and a structure-aware repulsion loss to push each pair of projections that fall inside the silhouette far away from each other. Although we omit surface interpolation, visibility handling, and shading, our results demonstrate that DRWR achieves state-of-the-art accuracies under widely used benchmarks, outperforming previous methods both qualitatively and quantitatively. In addition, our training times are significantly lower due to the simplicity of DRWR.
APA
Han, Z., Chen, C., Liu, Y. & Zwicker, M.. (2020). DRWR: A Differentiable Renderer without Rendering for Unsupervised 3D Structure Learning from Silhouette Images. Proceedings of the 37th International Conference on Machine Learning, in Proceedings of Machine Learning Research 119:3994-4005 Available from http://proceedings.mlr.press/v119/han20b.html.

Related Material