yrcong / STTran

Spatial-Temporal Transformer for Dynamic Scene Graph Generation, ICCV2021
MIT License
181 stars 34 forks source link

about object_bbox #40

Closed wusuowei11 closed 1 year ago

wusuowei11 commented 1 year ago

请问文件里面这个object_bbox标注位置信息的格式是什么,左上右下坐标(x1,y1,x2,y2)还是中心宽高坐标(x,y,w,h)呢

zee-fee commented 1 year ago

In Action Genome annotation, person_bbox has (xyxy) format and object boxes have (xywh) format. This is chnaged in the ActionGenome Dataset class, where all boxes are in xyxy format Hope this helps :)

wusuowei11 commented 1 year ago

感谢您的回复,我再确认一下,您是说object_bbox_and_relationship_filtersmall.pkl 和 object_bbox_and_relationship.pkl 中的object_bbox位置信息已经修改为(xyxy)格式了么

zee-fee commented 1 year ago

In object_bbox_and_relationship_filtersmall.pkl and object_bbox_and_relationship.pkl files, object bbox is in xywh format. It is converted to xyxy format in dataloader/action_genome.py line 133.

k['bbox'] = np.array([k['bbox'][0], k['bbox'][1], k['bbox'][0]+k['bbox'][2], k['bbox'][1]+k['bbox'][3]]) # from xywh to xyxy

wusuowei11 commented 1 year ago

感谢您的回复,我理解您的意思啦。在Action Genome annotation的(xywh)格式中,xy是指边框的左上角顶点坐标。

yrcong commented 1 year ago

感谢您的回复,我理解您的意思啦。在Action Genome annotation的(xywh)格式中,xy是指边框的左上角顶点坐标。

I remember, in the original Action Genome annotations, "person" boxes are annotated with xywh, while the boxes of other classes are annotated in xyxy.

zee-fee commented 1 year ago

感谢您的回复,我理解您的意思啦。在Action Genome annotation的(xywh)格式中,xy是指边框的左上角顶点坐标。

That's correct.

zee-fee commented 1 year ago

Here is a jupyter notebook where I explored the dataset. Hope you find this helpful. Notebook

wusuowei11 commented 1 year ago

非常感谢您,这对我理解帮助很大。