[논문 요약26] [9] P. Dollar, C. Wojek, B. Schiele, and P. Perona. Pedestrian detection: An evaluation of the state of the art. TPAMI, 34(4):743–761, 2012. 1, 2, 5, 6

[업데이트 2018.09.06 15:32]

1. 베이스 논문

[논문 요약17] Perceptual Generative Adversarial Networks for Small Object Detection

(http://arclab.tistory.com/172)

Extensive evaluations on the challenging Tsinghua-Tencent 100K [45] and the Caltech [9] benchmark well demonstrate the superiority of Perceptual GAN in detecting small objects, including traffic signs and pedestrians, over well-established state-of-the-arts.

We evaluate our Perceptual GAN method on the challenging Tsinghua-Tencent 100K [45] and the Caltech benchmark [9] for traffic sign and pedestrian detection respectively.

The Caltech benchmark [9] is the most popular pedestrian detection dataset. About 250,000 frames with a total of 350,000 bounding boxes and 2,300 unique pedestrians are annotated. We use dense sampling of the training data (every 4th frame) as adopted in [44, 27]. Following the conventional evaluation setting [9], the performance is evaluated on pedestrians over 50 pixels tall with no or partial occlusion, which are often of very small sizes. The evaluation metric is log-average Miss Rate on False Positive Per Image (FPPI) in [10􀀀2; 100] (denoted as MR following [42]).

For the Caltech benchmark [9], we utilize the ACF pedestrian detector [7] trained on the Caltech training set for object proposals generation.

Since the pedestrian instances on the Caltech benchmark [9] are often of small scales, the overall performance on it can be used to evaluate the capability of a method in detecting small objects.

2. 레퍼런스 논문

[9] P. Dollar, C. Wojek, B. Schiele, and P. Perona. Pedestrian detection: An evaluation of the state of the art. TPAMI, 34(4):743–761, 2012. 1, 2, 5, 6

(http://mlsurveys.s3.amazonaws.com/97.pdf)

3. 주요 내용 요약

3.1 레퍼런스 논문 주요 내용

보행자 인식하는 문제는 컴퓨터비전의 주요 해결 문제 중에 하나입니다. 그러나 여러가지 데이터 셋이 사용되고, 평가 방법이 제각각이라 성능에 있어서 비교가 어려운데, 본 논문에서는 평가 등에 활용 될 수 있도록 잘 정리된 보행자 데이터셋과 평가 방법을 제안합니다.

3.2 베이스 논문에서 인용한 내용

Caltech benchmark를 통해 학습 및 검증을 하며, 기존 방법들과 성능을 비교합니다.