Dataset Information
-
The basic annotation information on RAP dataset can be seen from the Table 1.
Table 1: Annotations in the RAP dataset Class Attribute Spatial-Temporal Time, SceneID, image position,Bounding box of body/head-shoulder/upper-body/lower-body/accessories. Whole Gender, age, body shape, role. Accessories Backpack, single shoulder bag, handbag, plastic bag, paper bag etc. Posture,Actions Viewpoints, telephoning, gathering, talking, pushing, carrying etc. Occlusions Occluded parts, occlusion types. Parts head Hair style, hair color, hat, glasses. upper Clothes style, clothes color. Lower Clothes style, clothes color, footware style, footware color.
-
To have a clear understanding about the annotation, some attributes are shown in Figure 1 according to the class which is shown in Table 1.
Figure 1: Attribute examples in RAP dataset. The attributes are shown according to defination in Table 1.
-
In real scene, attribute will change or be hard to determine due to camera viewpoint,
body part occlusion, human's orientation and pose, time range, image qualiry etc.
Even the same person's attribute will change a lot, which make it a challenge problem.
Examples of different attributes on the same person in RAP dataset are shown in Figure 2.
Figure 2: The same person which are captured at different time.
-
The RAP dataset provides rough annotation about body parts, including head-shoulder, uppper body and lower body. Accurate annotation on part position
is hard in that the image quality is not good in real surveillance scenarios. The part anontation examples are shown in Figure 3.
Figure 3: The part annotation example in RAP dataset.
- Viewpoints and Occlusion are two important research topics in visual recogniton tasks. The RAP dataset provides the annotation on these two factors. The distribution of viewpoints and occlusion positons in RAP are shown in Figure 4.
Figure 4: The distribution of viewpoints and occlusion positions.
- For clear comparsion with existing person attribute dataset, the detail information about existing popular pedestrian attribute dataset has been shown in Table 2.
Table 2: A comparison between different pedestrian attribute datasets Datasets #Cams Scene Annotation unit #Samples Resolution #Binary attributes Viewpoint Occlusion Part location VIPeR[1] 2 outdoor PID 1264 48*128 21 yes no no PRID[2] 2 outdoor PID 400 64*128 21 no no no GRID[3] 8 outdoor PID 500 from 29*67 to 169*365 21 no no no APiS[4] - outdoor PI 3661 48*128 11 no no no PETA[5] - mixture PID 19,000 from 17*39 to 169*365 61 no no no RAP 26 indoor PI 41,585 from 36*92 to 344*554 69 yes yes yes
- Viewpoints and Occlusion are two important research topics in visual recogniton tasks. The RAP dataset provides the annotation on these two factors. The distribution of viewpoints and occlusion positons in RAP are shown in Figure 4.