wangxun

最后一个等号后面的i ----> i+n的是目标函数J对x_{i+n} 求偏导数。 包括下面的那句话的i 也要换成i+n 

more conciser sampler

2

https://github.com/michuanhaohao/reid-strong-baseline/blob/master/data/samplers/triplet_sampler.py ```bash batch_idxs = [] for idx in idxs: batch_idxs.append(idx) if len(batch_idxs) == self.num_instances: batch_idxs_dict[pid].append(batch_idxs) batch_idxs = [] ``` can be replaced with ```bash batch_idxs_dict[label] = [idxs[i*self.num_instances : (i+1)*self.num_instances] for...

E2E-TAD code

2

Do you have a planed time to release the code of E2E-TAD?

Why stoping grad for the bias of projection on k?

1

https://github.com/microsoft/Swin-Transformer/blob/afeb877fba1139dfbc186276983af2abb02c2196/models/swin_transformer_v2.py#L149

when the code will be released?

two questions L2norm and zero losses

7

I have replicated your papar : > , on the Market1501 dataset by caffe. It has good performance just as your paper said. As I can see, the Batch hard...

The version of pytorch

1

Great work! I want to ask which version of pytorch to use. thx!

results in application is not good and some other problems

2

In experiment, I find ROLO is not better than YOLO, and the IOU is not low because the ROLO usually give a smaller bounding box. Secondly, in the ROLO project,...

How is your performance? Does it consistent with the paper?