3D reconstruction for ShapeNet dataset

数据集下载

bash download_dataset.sh

训练

git clone https://github.com/Jittor/shapenet-reconstruction-jittor.git
cd shapenet-reconstruction-jittor
git clone https://github.com/Jittor/jrender.git
mv jrender jrender-bak
mv jrender-bak/jrender .
python train.py -eid rcon

测试

python test.py -eid test -d 'data/results/models/recon/checkpoint_0250000.pkl'

速度测试

	200次训练所用时间
Jittor	15.66
PyTorch	19.25
加速比	1.22

指标对比

Class	IoU(Jittor)	IoU(pytorch)	IoU(paper with pytorch)
Airplane	0.651	0.650	0.642
Bench	0.501	0.488	0.508
Cabinet	0.697	0.688	0.711
Car	0.738	0.732	0.770
Chair	0.526	0.526	0.527
Display	0.627	0.610	0.616
Lamp	0.459	0.457	0.463
Loudspeaker	0.654	0.651	0.665
Rifle	0.686	0.688	0.681
Sofa	0.687	0.685	0.688
Table	0.458	0.459	0.449
Telephone	0.854	0.827	0.790
Watercraft	0.617	0.613	0.595
OverAll	0.627	0.621	0.623