search

autonomousvision / convolutional_occupancy_networks

[ECCV'20] Convolutional Occupancy Networks
https://pengsongyou.github.io/conv_onet
MIT License
830 stars 113 forks source link

How to train a new model for your own dataset #58

Open zhao-you-fei opened 1 year ago

zhao-you-fei commented 1 year ago

If I only have images and pointclouds, how to train a new model for my own data set

Barnonewdm commented 1 year ago

My suppose is to build the surface based on the point clouds.

zhao-you-fei commented 1 year ago

那如何用自己的点云数据训练一个 新的模型（train）不是generation或者test，我看每个数据集要求的格式是不一样的，config/default.yaml该如何配置

pengsongyou commented 1 year ago

This paper is a fully supervised approach, so you need to have watertight meshes in order to create occupancy as the supervision signal. If you only have point cloud, you need to find a way first to create a watertight mesh from it.

On Wed, 14 Dec 2022, 09:15 zhao-you-fei @.***> wrote:

那如何用自己的点云数据训练一个 新的模型（train）不是generation或者test，我看每个数据集要求的格式是不一样的，config/default.yaml该如何配置

— Reply to this email directly, view it on GitHub https://github.com/autonomousvision/convolutional_occupancy_networks/issues/58#issuecomment-1350603500, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADNJ3ITHCDLMAXLJAJPNHQ3WNF62RANCNFSM6AAAAAAS6EDFIM . You are receiving this because you are subscribed to this thread.Message ID: <autonomousvision/convolutional_occupancy_networks/issues/58/1350603500@ github.com>

zhao-you-fei commented 1 year ago

请问自己的点云数据生成的mesh如何评估，评估的是点云还是mesh，为什么我在代码中只看到读取的pointcloud.npz和points_iou.npz，没有真值怎么评估，包括提供的数据集我也没看到真值啊，既然要评估mesh不应该要有真值吗？