SPDA-CNN: Unifying Semantic Part Detection and Abstraction for Fine-Grained Recognition

Zhang, Han; Xu, Tao; Elhoseiny, Mohamed; Huang, Xiaolei; Zhang, Shaoting; Elgammal, Ahmed; Metaxas, Dimitris N.

doi:10.1109/cvpr.2016.129

Cited by 255 publications

(169 citation statements)

References 26 publications

Supporting

Mentioning

167

Contrasting

Unclassified

Order By: Relevance

“…Therefore, we integrate these 7 parts (beak, forehead, crown, nape, throat, left eyes and right eyes) into one "part", which is called "Head", and an identical label is assigned to the pair of legs and wings, respectively. We obtain 7 "parts", i.e., Head, Back, Belly, Breast, Leg, Wing and Tail, which is similar with [1]. We use the average of 6 "parts" (except the "Wings", as the result can be significantly influenced by the pose of the "Wings") as "Center" as the following function,…”

Section: Locationmentioning

confidence: 98%

“…On the other hand, as illustrated in the state-of-the-art methods [1], [3], [6], [7], bounding boxes are used in both training and testing phase, but part annotation is only used in training. In order to control variables, the part annotation should not be used in the testing step, so the "parts" should be detected first.…”

Section: Locationmentioning

confidence: 99%

“…In order to control variables, the part annotation should not be used in the testing step, so the "parts" should be detected first. After comparison with the other algorithms, we choose the algorithm proposed by H. Zhang [1] to implement the part detection. However, different "part" has different detection accuracy, so we can't treat them equally.…”

Section: Locationmentioning

confidence: 99%

“…1, the "parts" around the "Head" are so dense. If we use the traditional 1 4 Height× 1 4 Width regions [8] to represent these "parts" (where Height and Width represents the height and width of the bounding box, respectively), these regions will be highly overlapping. Therefore, we integrate these 7 parts (beak, forehead, crown, nape, throat, left eyes and right eyes) into one "part", which is called "Head", and an identical label is assigned to the pair of legs and wings, respectively.…”

Section: Locationmentioning

confidence: 99%

“…It has been proven in many literatures [3], [6] that using the features automatically extracted from CNN and classifying with SVM is a good scheme for classification task. Recently, a new framework, which is called "end to end system" [1], has been proposed. In this system, a Neural Network is constructed and trained.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Pre-Processing for Fine-Grained Image Classification

Yang

et al. 2017

IEICE Trans. Inf. & Syst.

View full text Add to dashboard Cite

SUMMARYRecently, numerous methods have been proposed to tackle the problem of fine-grained image classification. However, rare of them focus on the pre-processing step of image alignment. In this paper, we propose a new pre-processing method with the aim of reducing the variance of objects among the same class. As a result, the variance of objects between different classes will be more significant. The proposed approach consists of four procedures. The "parts" of the objects are firstly located. After that, the rotation angle and the bounding box could be obtained based on the spatial relationship of the "parts". Finally, all the images are resized to similar sizes. The objects in the images possess the properties of translation, scale and rotation invariance after processed by the proposed method. Experiments on the CUB-200-2011 and CUB-200-2010 datasets have demonstrated that the proposed method could boost the recognition performance by serving as a pre-processing step of several popular classification algorithms.

show abstract

Section: Locationmentioning

confidence: 98%