| Summary: | The self-supervised monocular depth estimation paradigm has become an important branch of computer vision depth-estimation tasks. However, the depth estimation problem arising from object edge depth pulling or occlusion is still unsolved. The grayscale discontinuity of object edges leads to a relatively high depth uncertainty of pixels in these regions. We improve the geometric edge prediction results by taking uncertainty into account in the depth-estimation task. To this end, we explore how uncertainty affects this task and propose a new self-supervised monocular depth estimation technique based on multi-scale uncertainty. In addition, we introduce a teacher–student architecture in models and investigate the impact of different teacher networks on the depth and uncertainty results. We evaluate the performance of our paradigm in detail on the standard KITTI dataset. The experimental results show that the accuracy of our method increased from <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><mn>87.7</mn><mo>%</mo></mrow></semantics></math></inline-formula> to <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><mn>88.2</mn><mo>%</mo></mrow></semantics></math></inline-formula>, the <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><mi>A</mi><mi>b</mi><mi>s</mi><mi>R</mi><mi>e</mi><mi>l</mi></mrow></semantics></math></inline-formula> error rate decreased from <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><mn>0.115</mn></mrow></semantics></math></inline-formula> to <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><mn>0.11</mn></mrow></semantics></math></inline-formula>, the <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><mi>S</mi><mi>q</mi><mi>R</mi><mi>e</mi><mi>l</mi></mrow></semantics></math></inline-formula> error rate decreased from <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><mn>0.903</mn></mrow></semantics></math></inline-formula> to <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><mn>0.822</mn></mrow></semantics></math></inline-formula>, and the <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><mi>R</mi><mi>M</mi><mi>S</mi><mi>E</mi></mrow></semantics></math></inline-formula> error rate decreased from <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><mn>4.863</mn></mrow></semantics></math></inline-formula> to <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><mn>4.686</mn></mrow></semantics></math></inline-formula> compared with the benchmark Monodepth2. Our approach has a positive impact on the problem of texture replication or inaccurate object boundaries, producing sharper and smoother depth images.
|
|---|