Revisiting Single Image Depth Estimation: Toward Higher Resolution Maps With Accurate Object Boundaries

By Junjie Hu, Mete Ozay, Yan Zhang, Takayuki Okatani. IEEE Workshop/Winter Conference on Applications of Computer Vision, 2018

425 citations recorded by the source corpus.

This paper considers the problem of single image depth estimation. The employment of convolutional neural networks (CNNs) has recently brought about significant advancements in the research of this problem. However, most existing methods suffer from loss of spatial resolution in the estimated depth maps; a typical symptom is distorted and blurry reconstruction of object boundaries. In this paper, toward more accurate estimation with a focus on depth maps with higher spatial resolution, we propose two improvements to existing approaches. One is about the strategy of fusing features extracted at different scales, for which we propose an improved network architecture consisting of four modules: an encoder, decoder, multi-scale feature fusion module, and refinement module. The other is about loss functions for measuring inference errors used in training. We show that three loss terms, which measure errors in depth, gradients and surface normals, respectively, contribute to improvement of accuracy in an complementary fashion. Experimental results show that these two improvements enable to attain higher accuracy than the current state-of-the-arts, which is given by finer resolution reconstruction, for example, with small objects and object boundaries.

Subfields: Computer Vision and Pattern Recognition, Media Technology

View Paper: https://www.semanticscholar.org/paper/034eff6fda29c70c861ffac18882ea129249555f

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Media Technology

Citations in the corpus

Delving into Deep Imbalanced Regression (2021, positive): “Following (Hu et al., 2019), for both training and evaluation phases, we ﬁrst downsample images (both RGB and depth) from original size 640 × 480 to 320 × 240 using bilinear interpolation, then conduct center crop to obtain images of size 304 × 228 , and ﬁnally normalize them into [0 , 1] .…”
Enforcing Geometric Constraints of Virtual Normal for Depth Prediction (2019, methodology): “By contrast, although the previously state-of-the-art method [21] predicts the depth with low errors, the reconstructed point cloud is far away from the original shape (see, e.”
GeoWizard: Unleashing the Diffusion Priors for 3D Geometry Estimation from a Single Image (2024, methodology): “Some existing approaches propose to explicitly acquire the surface normal from the depth map by using some geometric constraints, such as Sobel-like operator [19, 27], differentiable least square [36, 44], or randomly sampled point triplets [37, 69, 70].”
AdaBins: Depth Estimation Using Adaptive Bins (2020, methodology): “Monocular depth estimation has been considered by many CNN methods as a regression of a dense depth map from a single RGB image [8, 25, 45, 16, 46, 11, 19, 1, 26, 22].”
Self-Supervised Monocular Depth Hints (2019, background): “We found that l1 s works well with estimated depths (and [15] favors it for Kinect data), while l s is an established choice for accurate LiDAR and SLAM depths [18, 19] and disparities [35].”
Clear Grasp: 3D Shape Estimation of Transparent Objects for Manipulation (2019, background): “More recent works study how to obtain 3D data from color images by directly inferring depth images [14, 30, 10, 41, 22, 52, 16], or ﬁlling in missing depth values in RGB-D images captured by commodity 3D cameras [21, 17, 59].”
ZoeDepth: Zero-shot Transfer by Combining Relative and Metric Depth (2023, neutral): “[14] SENet-154 157M Chen et al .”
Benchmarking Classic and Learned Navigation in Complex 3D Environments (2019, methodology): “For monocular depth estimation, we use the deep model of Hu et al. [19].”