Mingzhe comments

Results 23 comments of


                                            Mingzhe

General Inquiry on the GNN Reranking for Distance Calculation

Hi, thanks for the reply. What do you mean by mentioning transferring cosine similarity to euclidean distance in the reranking code? I read your paper and cosine distance is always...

General Inquiry on the GNN Reranking for Distance Calculation

Thanks for the reply. Maybe it will be better to try to transform the cosine similarity to euclidean distance first. BTW, do we need an additional square operation [here](https://github.com/layumi/Person_reID_baseline_pytorch/blob/master/re_ranking.py#L48)?

Question about FPS

Does the tracking speed have something to do with the detector? I tried with YoloV5 and got a faster tracking result.

Model Differences From The Reference Paper

Same question here... I also don't know why we need a classifier after having n-dim features for matching.

Integrating a new Re-ID model

I am thinking of the embedding dim for the Reid model. Since the mainstream model is ResNet-50, is it okay to have a 2048-dim vector as feature? (However we only...

只给了一个train.json文件，怎么 generate train.txt and test.txt？

The URL is expired...

CrossEntropyLoss reduction write error

@AutumnZ-94 Try this. They should be the same ```python x = torch.randn(32, 64) y1 = x.mean(0).sum() y2 = x.sum(1).mean() assert torch.allclose(y1, y2) ```

Pre-trained weights for visual and text adapter.

Inconsistent loss function from the paper?

Does it mean the loss is actually has nothing to do with the label? Like -1 for a fake image.

Inconsistent loss function from the paper?

> > after debug a lot I found this code is not wrong, but confusing > > ```python > > errD_real = netD(inputv) > > errD_real.backward(one) > > > >...