3D Shape Segmentation via Shape Fully Convolutional Networks

paper

Enviroment: 201610 version of caffe

caffe201610

Data

plier_fea

1.segmesh

to calculate distance between two neighbour meshes.

2.preprocessing (you need to run the python files to build the data for SFCN)

requirement: python 2.7, numpy, scipy, sklearn

a.edit the all_cimbine.py

--old_feature_dir : 3d shapes' feature files

--dir_all: 3d shapes' other files(dist, area, adj, seg)

b.set the output dir

--dir_output

c.run

Ps: More details are in ./preprocessing/README.md

3.build caffe

see caffe_read.txt

4.caffe-plier

a.run gen_lmdb.py to generate the lmdb dataset for caffe

b.run eval-solve.py

Ps: More details are in ./caffe-plier/README.md

5.SFCN's multilabel graph cut

if you want to run graph cut, you need multilabel graph cut code(you can find it in gco)