Multi-Scale Fusion Network Using Adaptive Cost Volume Filtering for Stereo Matching
- 주제(키워드) Stereo Matching
- 주제(DDC) 006.31
- 발행기관 아주대학교
- 지도교수 허용석
- 발행년도 2022
- 학위수여년월 2022. 8
- 학위명 석사
- 학과 및 전공 일반대학원 인공지능학과
- 실제URI http://www.dcollection.net/handler/ajou/000000032262
- 본문언어 영어
- 저작권 아주대학교 논문은 저작권에 의해 보호받습니다.
초록/요약
While recent deep learning-based stereo matching networks have shown outstanding advances, there are still some unsolved challenges. First, most state-of-the-art stereo models employ 3D convolutions for 4D cost volume aggregation,which limit the deployment of networks for resource-limited mobile environments owing to heavy consumption of computation and memory. Second, most stereo networks indirectly supervise cost volumes through disparity regression loss by using the softargmax function. This causes problems in ambiguous regions, such as the boundaries of objects, because there are many possibilities for unreasonable cost distributions which result in overfitting problem. To address these problems, we first propose an efficient multi scale sequential feature fusion network (MSFFNet). Specifically, we connect multi-scale SFF modules in parallel with a cross-scale fusion function to generate a set of cost volumes with different scales. These cost volumes are then effectively combined using the proposed interlaced concatenation method. Second, we propose an adaptive cost volume filtering (ACVF) loss function that directly supervises our estimated cost volume. The proposed ACVF loss directly adds constraints to the cost volume using the probability distribution generated from the ground truth disparity map and that estimated from the teacher network with higher accuracy. Results of several experiments using representative datasets for stereo matching shows that our proposed method is more efficient than previous methods. Concretely, our network architecture consumes fewer parameters and generates reasonable disparity maps with faster speed compared with the existing state-of-the art stereo models.
more목차
1 Introduction 1
2 Related Works 7
2.1 Efficient Stereo Matching Networks 7
2.2 Cost Volume Filtering 9
3 Proposed Method 11
3.1 Feature Extractor 12
3.2 Cost Aggregation 12
3.3 Interlaced Cost Volume 17
3.4 Disparity Regression and Refinement 17
3.5 Loss Function 19
3.5.1 Disparity Regression Loss 19
3.5.2 Adaptive Cost Volume Filtering Loss 19
3.5.3 Total Loss Function 23
4 Experiments and Results 24
4.1 Datasets and Evaluation Metrics 24
4.2 Implementation Details 25
4.3 Comparative Result 26
4.3.1 SceneFlow Result 26
4.3.2 KITTI 2015 Result 27
4.4 Ablation Study 30
4.4.1 Effect of Interlaced Cost Concatenation 30
4.4.2 Effect of Adaptive Cost Volume Filtering Loss 32
5 Conclusions 33
References 34
