chun

Results 26 issues of chun

Hi, thanks for your great work. I want to use these IoU losses on KITTI and evaluate the model performance. How should I do, and how to write the transform...

作者好,我想用你的方法来训练自己的数据,碰到以下两个问题,想请教下: 1. 在训练过程中,proposal_target_layer_cascade.py中类方法_ProposalTargetLayer下的bg_num_rois和fg_num_rois这两个参数经常出现异常而导致训练终端,即二者都等于0,出现raise错误。这可能是什么原因呢?此外,我检查了这两个参数的计算由来,是通过max_overlap结果与所设置阈值比较来决定是fg还是bg,这部分同样存在一个问题就是cfg.TRAIN.BG_THRESH_LO这个参数,无论我在config.py文件中如何设置,打印出来的值始终为0.0,这是什么原因呢。此外,还让我不解的是,这部分代码写的是当max_overlap值处于cfg.TRAIN.BG_THRESH_LO和cfg.TRAIN.BG_THRESH_HI之间时判定为bg样本,即0.0~0.5认为是bg。那我上诉的第一个问题,在实际训练过程出现fg_num_rois和bg_num_rois都为0的情况就变得很矛盾了,请问你有什么好的建议和想法吗,我困惑了很久。 2. 对于训练batch_size的设置,代码默认是1,请问只能设置为1吗?我有两个疑惑,首先设置为1每次只喂给模型1个样本,这样训练出来的模型是否会有泛化能力呢?(按以往的经验,应该是batsh size越大训练出来的模型效果越好,而且1的设置有些奇怪),另外就是当我修改这个bs参数,例如bs=16,这时会报错:这个错误位置是出现在resnet_HTCN.py中,在self.RandomLayer这部分的矩阵计算会出现前后维度不匹配,这个不匹配体现在某一维度的大小是原来大小与现在bs=16的乘积,请问这部分的错误是什么原因呢?(如果bs=1则不会有这个问题,乘上1还是本身的维度大小) 非常期待你的答复,为我解答疑惑,感谢!

Hi, could you please share the data on Baiduyun? The Google Drive is unstable, and the data downloading often interrupt

Hi, author. I want to generate my own simulated V2X data, how should I step-by-step?

Hi, thanks for your great work. If I want to test an input video from the camera, how should I do?

Hi, author. It is a great data visualization job, but I confuse how to use it to visualize my data or detection result? Just git clone it in my project?...

什么代码,运行完都没东西!

Hi, author. Thanks for your great jobs. 1. When does the project support for the other datasets as described in the paper, i.e., V2X-SIm and Coperception? 2. How to access...

Hi, author. I have downloaded the HEVI dataset, and prepare to train the model. But I am confused about the dataset definition in the config.yaml. For instance, in 'fol_ego_train.yaml', what...