search

Dawn-LX / OpenVoc-VidVRD

Official code for the ICLR2023 paper Compositional Prompt Tuning with Motion Cues for Open-vocabulary Video Relation Detection
37 stars 1 forks source link

object tracking代码 #1

Closed Iriya99 closed 11 months ago

Iriya99 commented 1 year ago

请问object tracking部分是没有公布代码吗？tracklet是用什么模型提取的呢？

Dawn-LX commented 1 year ago

we use Seq-NMS to build object tracklets, refer to https://github.com/tmoopenn/seq-nms

As for how to moidify the code to VidVRD setting, we follow the code of this paper "Video Visual Relation Detection via Iterative Inference ACM MM 2021" (shangxin@comp.nus.edu.sg)

Iriya99 commented 1 year ago

How can I test the model on my own video?

Dawn-LX commented 1 year ago

How can I test the model on my own video?

What do you mean? test only the quality of object tracklets ?

Dawn-LX commented 11 months ago

Please refer to this repo VidSGG-TrajDataPrepare for how to prepare the above traj data, including 1)detect bbox --> 2)object tracking --> 3)extract RoI features.

Iriya99 commented 8 months ago

Thanks very much.