EveningLin

The code does not seem to be consistent with the formula given in 3.3

2

``` if self.args.with_reid: emb_dists = matching.embedding_distance(strack_pool, detections) / 2.0#计算特征的cost矩阵 raw_emb_dists = emb_dists.copy() emb_dists[emb_dists > self.appearance_thresh] = 1.0 emb_dists[ious_dists_mask] = 1.0 dists = np.minimum(ious_dists, emb_dists)#使用矩阵中每个元素的最小值作为成本矩阵C的最终值 ``` （1）the first question is Why...

【cliff】Human3.6M的权重能不能release一下

1

在做annotation,但是官方realse的权重在human36M上都表现都比较一般，测出来的mpjpe太大了不适合进一步做数据集，希望能realse一下human36的权重

【Cliff】annotation on human36M

6

我使用cliff作为annotator，参考EFT的架构，将原先eft的部分更换成了cliff，但是效果上测出来整体的mpjpe在70左右，而且非常耗时，仅使用s9和s11,耗时接近20个小时。能否将你们的annotator的代码release一下或者是给出一些指导性意见。

Thank you very much for your work. Here are some questions I hope you can answer (1)What are the meanings of the red boxes  (2) why we should merge...

Why perform such operations on k and v as shown in the above diagram?

1

 Why perform such operations on k and v as shown in the above diagram?

 请问这里为什么是3和8呢

NMS not in the c++ api

2

make it not so efficient when dealing with Postprocessing in c++