ColTrack
ColTrack copied to clipboard
bytedance
This repository is an official implementation of Collaborative Tracking Learning for Frame-Rate-Insensitive Multi-Object Tracking
您好，我在执行第一阶段训练的test阶段时，`track_core`中有呼叫`build_tracker`函数https://github.com/bytedance/ColTrack/blob/c99e636ddaa6ddea6b50e02630bd80c830319392/motlib/tracker/offline_track.py#L12-L13 但`args.tracker_name`如果使用预设的`ColTrack`，会回传没有注册`ColTrack`的错误，**实际trace程式码也并没有找到`ColTrack`类别**，只有`ByteTracker`与`BotTracker` 有进行`@TRACKER_REGISTRY.register() `。 目前我的解决方式是把`args.tracker_name`设置成`ByteTracker` 以下是我的疑問: 1. 请问在evaluation阶段，[track_by_offline()](https://github.com/bytedance/ColTrack/blob/c99e636ddaa6ddea6b50e02630bd80c830319392/motlib/evaluation/evaluators/my_coco.py#L208) 是负责执行`ColTrack`以外的**TrackByDetection**演算法 吗? 目的是用来与ColTrack比较性能? 2. 实际上要得到ColTrack的HOTA, MOTA等分数应该也是使用evaluation的code，如果没有ColTrack的Tracker，我们要如何进行模型的Inference呢?
Hello~ Thank you for your amazing paper. I would like to ask about **the input of IRM**. In Figure 3, the input is described as $X^t=(F^t, \hat{B^t})$. but in Equation...
Hi, when training the models reported in table 4,5,6, is the downsampling interval of datasets set as 1 or other numbers?
Hello, I researched Multi-Object Tracking with query in South Korea. Thanks for your wonderful works. I want to train your model on bdd100k. but your readme file didn't include it....
