TinyZeaMays / CircleLoss

Pytorch implementation of the paper "Circle Loss: A Unified Perspective of Pair Similarity Optimization"

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
Updated 4 months ago1
self.soft_plus(torch.logsumexp(logit_n, dim=0) + torch.logsumexp(logit_p, dim=0))
Updated a year ago2
What if similarity_matrix[positive_matrix] is empty?
Updated 2 years ago
梯度图怎样画的？
Updated 2 years ago
circle loss 的适用范围？
Updated 2 years ago2
不收敛啊，需要PK采样么
Updated 2 years ago
RuntimeError: "triu" not implemented for 'Bool'
Updated 2 years ago8
Why do a_p and a_n use .detach()?
Closed 3 years ago
能用在语义分割中吗
Updated 3 years ago
The meaning of detaching sp and sn from the gradient gragh?
Updated 3 years ago
这个CircleLoss是基于批量数据实现？
Closed 4 years ago5
why are sp and sn both decreasing during training process?
Updated 4 years ago3
AttributeError: 'Tensor' object has no attribute 'logical_not'
Updated 4 years ago2
Which circle loss in your repo (circle_loss or circle_loss_early) consist with the paper
Closed 4 years ago1
bug?
Closed 4 years ago4
请问这个circle loss可以使用 sigmoid logits作为输入嘛
Closed 4 years ago1
关于circle loss代码实现中的梯度问题
Closed 4 years ago2
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
Updated 4 years ago20
Why the circle loss's output is 0?/ 为什么loss的输出是0？
Closed 4 years ago3
有人遇到这个问题吗 AttributeError: 'Tensor' object has no attribute 'logical_not'
Closed 4 years ago
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？
Updated 4 years ago27
softplus failed to compute its gradient
Closed 4 years ago1
pair方式的metric
Closed 4 years ago3
label_matrix.triu(diagonal=1) gpu和cpu运行区别
Closed 4 years ago1
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
Closed 4 years ago5
AttributeError: module 'torch' has no attribute 'clamp_min'
Closed 4 years ago1
Could you show us the dataset's format？
Closed 4 years ago4
请问circle_loss.py中最后要用softplus操作？
Closed 4 years ago2
How to obtain class labels?
Closed 4 years ago1
RuntimeError: name not implemented for 'Bool'
Closed 4 years ago2
Pair-wise level or Class-level?
Closed 4 years ago6
直接忽略batch维度相加sn/sp似乎不太对
Closed 4 years ago4
Why is the CircleLossLikeCE loss function under the mnist_example.py?
Closed 4 years ago4
Are you sure that your code was consistent with this paper?
Closed 4 years ago5
to fix bug
Closed 4 years ago