TinyZeaMays/CircleLoss

Issues

• 我想用在基于CNN的实例级图像检索中去，请问能够实现吗

  #35 opened 2 years ago by huanhao15
  1

• self.soft_plus(torch.logsumexp(logit_n, dim=0) + torch.logsumexp(logit_p, dim=0))

  #30 opened 4 years ago by Heath-zyl
  2

• What if similarity_matrix[positive_matrix] is empty?

  #34 opened 3 years ago by liamsun2019
  0

• 梯度图怎样画的？

  #33 opened 3 years ago by genhao3
  0

• circle loss 的适用范围？

  #18 opened 5 years ago by Singapore-mor
  2

• 不收敛啊，需要PK采样么

  #32 opened 3 years ago by onlinehuazai
  0

• RuntimeError: "triu" not implemented for 'Bool'

  #17 opened 5 years ago by PR1706
  8

• Why do a_p and a_n use .detach()?

  #31 opened 4 years ago by zs-zhong
  0

• 能用在语义分割中吗

  #29 opened 4 years ago by feivelliu
  0

• The meaning of detaching sp and sn from the gradient gragh?

  #28 opened 4 years ago by WeiQiangRao
  0

• 这个CircleLoss是基于批量数据实现？

  #22 opened 4 years ago by L-lei
  5

• why are sp and sn both decreasing during training process?

  #27 opened 4 years ago by wj1tr0y
  3

• AttributeError: 'Tensor' object has no attribute 'logical_not'

  #24 opened 4 years ago by kevinbro96
  2

• Which circle loss in your repo (circle_loss or circle_loss_early) consist with the paper

  #26 opened 4 years ago by vodanhbk95
  1

• bug?

  #25 opened 4 years ago by interestingzhuo
  4

• 请问这个circle loss可以使用 sigmoid logits作为输入嘛

  #23 opened 4 years ago by y6216886
  1

• 关于circle loss代码实现中的梯度问题

  #12 opened 5 years ago by stupiding
  2

• 和arcface这些损失函数一样，circle loss也应该是和全连接层的权重求相似度吧，convert_label_to_similarity这个函数看不懂啊

  #6 opened 5 years ago by westnight
  20

• Why the circle loss's output is 0?/ 为什么loss的输出是0？

  #20 opened 4 years ago by Pengyang233
  3

• 有人遇到这个问题吗 AttributeError: 'Tensor' object has no attribute 'logical_not'

  #21 opened 4 years ago by silvercherry
  0

• 我使用了该loss，但在行人reid任务上，最终的性能很难达到作者公布的结果，请问作者有复现出这个结果吗？

  #8 opened 5 years ago by qiguming
  27

• softplus failed to compute its gradient

  #19 opened 5 years ago by LiuJMzzZ
  1

• pair方式的metric

  #16 opened 5 years ago by ruiming46zrm
  3

• label_matrix.triu(diagonal=1) gpu和cpu运行区别

  #15 opened 5 years ago by AaboutL
  1

• 想用circleloss替换facenet的tripletloss，tripletloss中的a,p,n通过embeding vector得到，但是有点不明白convert_label_to_similarity怎么实现sp,sn的

  #11 opened 5 years ago by NNUCJ
  5

• AttributeError: module 'torch' has no attribute 'clamp_min'

  #4 opened 5 years ago by guanyonglai
  1

• Could you show us the dataset's format？

  #5 opened 5 years ago by Dr-Zhou
  4

• 请问circle_loss.py中最后要用softplus操作？

  #10 opened 5 years ago by CloudHerods
  2

• How to obtain class labels?

  #9 opened 5 years ago by AddASecond
  1

• RuntimeError: name not implemented for 'Bool'

  #13 opened 5 years ago by buer0819
  2

• Pair-wise level or Class-level?

  #14 opened 5 years ago by haohang96
  6

• 直接忽略batch维度相加sn/sp似乎不太对

  #7 opened 5 years ago by PistonY
  4

• Why is the CircleLossLikeCE loss function under the mnist_example.py?

  #3 opened 5 years ago by JasonChenhx
  4

• Are you sure that your code was consistent with this paper?

  #2 opened 5 years ago by eeric
  5

• to fix bug

  #1 opened 5 years ago by TinyZeaMays
  0