Number of training samples used in paper

carlvinson70 commented 1 year ago

Hi there,

In the paper it says '~42k' raw KITTI samples are used - is this a typo? Apparently you only used ~18k train + eigen clean to get those results, or I am missing something here?

Hope you can clarify a bit. Thanks!

Cc-Hy commented 1 year ago

Hi, Sorry for being late, the best results in the paper use the entire kitti raw with ~42k samples, and the validation settings in this repo use ~18k train + eigen clean data. If you want to use the entire kitti raw, please use this split file and generate the label file for training.

xiaoxusanheyi commented 12 months ago

您好，抱歉迟到了，本文中的最佳结果使用了整个 kitti raw 和 ~42k 样本，并且此存储库中的验证设置使用 ~18k 训练 + 特征干净数据。如果你想使用整个kitti raw，请使用这个分割文件并生成标签文件进行训练。

你好作者，这个实验的话 kitti-3D数据集使用那个trainval(~7K)训练的结果没有给出吗？而表1中的KITTI-3D的测试结果是用~18k 训练 + 特征干净数据训练(如下图pkl文件) ，然后在测试集7518上进行测试得到的结果吗

Cc-Hy / CMKD

Number of training samples used in paper #43