Zhongdao/gcn_clustering

Issues

• why first iteration not use thershold in test code

  #45 opened 10 months ago by yongqis
  1

• about the features.zip

  #32 opened 4 years ago by KosukeHao
  3

• 计算边分数时， score_dict[e[0], e[1]] = 0.5*(score_dict[e[0], e[1]] + score[i]),这样会导致后遍历到的边分数权重更大，为什么不用真正的平均呢

  #28 opened 4 years ago by zhufq00
  1

• knn上界已经复现了，但我认为graph_propagation的max_sz不太合理，完全可以改成th逐渐递减的模式去划分clusters

  #44 opened 3 years ago by FDuiv
  1

• 无监督还是有监督？

  #35 opened 4 years ago by littlejiajia
  3

• bcubed函数里面，根据我的验证，precision和recall两个变量可能搞反了

  #43 opened 3 years ago by FDuiv
  0

• 模型测试效果不理想

  #11 opened 5 years ago by lichanglin3445667
  19

• face images of features

  #42 opened 4 years ago by mensaochun
  0

• Extracting features from ArcFace

  #33 opened 4 years ago by saedr
  2

• Images of features.zip

  #24 opened 4 years ago by ZPdesu
  1

• IJB-B list

  #41 opened 4 years ago by houguanqun
  0

• 关于原始数据集

  #40 opened 4 years ago by StalkerDAs
  0

• Using code in the new Gpu-s(Python Versions)

  #39 opened 4 years ago
  0

• usinng mean aggregation can not satisfy the requriment that G = g(X, A) is an aggregation matrix of size N × N and each row is summed up to 1

  #37 opened 4 years ago by fenglv12345
  0

• 关于特征提取

  #30 opened 4 years ago by zhufq00
  1

• About the dataset

  #36 opened 4 years ago by KimKeepLearning
  0

• 请问为什么训练的时候没有softmax，而测试的时候使用

  #31 opened 4 years ago by KimKeepLearning
  0

• 关于labels

  #34 opened 4 years ago by csqmaurine
  0

• features.zip

  #3 opened 5 years ago by xxx2974
  3

• 关于梯度回传

  #12 opened 5 years ago by lycwgf
  7

• 请问，GCN 如何使用 batch 训练？这里的batch 是指的 对 node的 分批训练吗？

  #26 opened 4 years ago by Emma0118
  1

• 我使用新数据集时，每个batch正样本比例只有1%左右，我使用resnet50提取特征，请问这有什么解决办法吗

  #29 opened 4 years ago by zhufq00
  0

• 想请问下，我没有改动代码任何地方，但是训练2个batch后precison和recall就变为0了,请问该如何训练出best.ckpt的效果呢

  #27 opened 4 years ago by zhufq00
  0

• 某些batch速度比较慢，有没有类似问题的啊？

  #10 opened 5 years ago by CaiQiuYu
  8

• Can the author provide the data preprocessing script

  #25 opened 4 years ago by angelhunt
  0

• 索引越界问题！

  #13 opened 5 years ago by xxx2974
  0

• About features of CASIA

  #8 opened 5 years ago by LZDSJTU
  14

• 运算矩阵大小

  #23 opened 5 years ago by ltm920716
  0

• the idea about GCN

  #22 opened 5 years ago by ltm920716
  1

• is this a clustering model or classification model？ why the code "self.labels = np.load(label_path)" is in feeder.py

  #21 opened 5 years ago by fusionchan
  0

• how to generate the KNN graph？

  #2 opened 5 years ago by whxhngns200
  4

• What is the ideal number of neighbour for any dataset?

  #20 opened 5 years ago by abhiksark
  0

• 关于邻接矩阵的变换

  #19 opened 5 years ago by yingjianling
  0

• 关于idea

  #18 opened 5 years ago by BluebirdStory
  1

• pred will contain nan if k_at_hop is large

  #6 opened 5 years ago by luzai
  4

• 测试的一个小 bug

  #17 opened 5 years ago by ludics
  0

• Ms-Celeb-1M的测试结果是多少，不知道作者有单独测试过吗

  #16 opened 5 years ago by Weizhongjin
  0

• 关于Pression和Recall的问题，代码中应当是计算反了

  #15 opened 5 years ago by CaiQiuYu
  1

• About acknowledgement

  #14 opened 5 years ago by XiaohangZhan
  1

• predicting edges are all zeros during training

  #7 opened 5 years ago by jxyecn
  2

• about training error.

  #9 opened 5 years ago by xxx2974
  0

• How to generate "attention aggregation" feature?

  #5 opened 5 years ago by jxyecn
  3

• Why use the random subset of CASIA?

  #4 opened 5 years ago by LZDSJTU
  1

• Could u pls provide the data preprocessing code?

  #1 opened 5 years ago by abcdvzz
  1