Multilabel single bounding box

[abhigoku10]

@DongDem @Purkialo hi thanks for opensourcing the code base, i wanted to know if we can extend this architecture where we can single bounding box detecting a person and then an additional head which gives multilabel classification ie gender age clothes ? is this possible to modify the network

Pls share your thought and thanks in advance

[yexiguafuqihao]

It is a good idea to increase more information to address pedestrian detection. An extra layer can be attached to each detection head. But sometimes it will introduce some negative effects into the detector. One instance can't be recognized if his/her head is occluded by the others when he/she is in uniform clothes, while is ubiquitous in the CrowdHuman dataset.