Abstract | ||
---|---|---|
Keypoint detection can help fine-grained bird recognition by aligning the bird with the detected keypoints. Most of the existing bird keypoint detection methods have poor performance on symmetric keypoints, because they mainly use texture features only, which usually can not distinguish between symmetric keypoints, such as the keypoints on left and right legs. Besides, these methods cannot deal well with the complex image background. Therefore, we propose a two-branch keypoint detection network that combines both 2D texture and 3D geometric features to tackle these problems. In the 2D branch, we use anchor loss to distinguish between foreground and background to alleviate the influence of complex background on keypoint detection. In the 3D branch, we introduce a 3D deformable mesh model to provide geometric information of symmetric keypoints. The prediction results of the two branches are fused to obtain the final keypoint detection results. We demonstrate the effectiveness of our proposed method on the widely-used CUB200-2011 [23] dataset. The experimental results show that our method can achieve superior accuracy in comparison with the state-of-the-art approaches. |
Year | DOI | Venue |
---|---|---|
2021 | 10.1007/978-3-030-87361-5_6 | IMAGE AND GRAPHICS (ICIG 2021), PT III |
Keywords | DocType | Volume |
Bird keypoint detection, Symmetric keypoint, 3D geometric information | Conference | 12890 |
ISSN | Citations | PageRank |
0302-9743 | 0 | 0.34 |
References | Authors | |
0 | 3 |
Name | Order | Citations | PageRank |
---|---|---|---|
Tingting Zhang | 1 | 0 | 0.34 |
Qijun Zhao | 2 | 419 | 38.37 |
Pubu Danzeng | 3 | 0 | 0.34 |