注册
登录
深度学习
>>
gvcnn-tf
>>
返回
项目作者:
ace19-dev
项目描述 :
Group-View Convolutional Neural Networks for 3D Shape Recognition
高级语言:
Python
项目主页:
项目地址:
git://github.com/ace19-dev/gvcnn-tf.git
创建时间:
2018-11-26T09:46:50Z
项目社区:
https://github.com/ace19-dev/gvcnn-tf
开源协议:
Apache License 2.0
下载
GVCNN (Group-View Convolutional Neural Networks for 3D Shape Recognition)
Data
Download
modelnet10-Class Orientation-aligned Subset
make .png images in order below
data_utils/make_views_dir.py
data_utils/off2obj.py (after ‘sudo apt install openctm-tools’)
data_utils/obj2png.py
Or You can create 2D dataset from 3D objects (.obj, .stl, and .off), using
BlenderPhong
.
Or Downsized modelnet40(from
https://drive.google.com/file/d/0B4v2jR3WsindMUE3N2xiLVpyLW8/view
) to modelnet12/6-view.
Quick Start
make group-view image tfrecord file
dataset_tools/create_modelnet_tf_record.py
train.py
TODO
balanced sampler
References from
http://openaccess.thecvf.com/content_cvpr_2018/papers/Feng_GVCNN_Group-View_Convolutional_CVPR_2018_paper.pdf
https://github.com/WeiTang114/MVCNN-TensorFlow
https://github.com/pclausen/obj2png