xy-guo / MVSNet_pytorch

PyTorch Implementation of MVSNet
619 stars 93 forks source link

代码中的matlab问题 #54

Open lgc1030 opened 1 year ago

SHERRYMINE commented 1 year ago
  1. Change the path parameters in the code
  2. Run BaseEvalMain_web.m
  3. Run ComputeStat_web.m
SHERRYMINE commented 1 year ago

In BaseEvalMain_web.m: 1.dataPath: download Points(6.3G) from http://roboimagedata.compute.dtu.dk/?page_id=36 2.plyPath: Mvsnet output path 3.resultsPath: BaseEvalMain_web.m output path

In ComputeStat_web.m: 1.dataPath: keeps the same path in BaseEvalMain_web.m 2.resultsPath:keeps the same path in BaseEvalMain_web.m

lgc1030 commented 1 year ago

In BaseEvalMain_web.m: 1.dataPath: download Points(6.3G) from http://roboimagedata.compute.dtu.dk/?page_id=36 2.plyPath: Mvsnet output path 3.resultsPath: BaseEvalMain_web.m output path

In ComputeStat_web.m: 1.dataPath: keeps the same path in BaseEvalMain_web.m 2.resultsPath:keeps the same path in BaseEvalMain_web.m

您好,我跑ComputeStat_web.m的结果是mean acc 0.4866 mean comp 0.3943,请问这个结果是论文中提到的最后结果吗?我不太理解ComputeStat_web.m和BaseEvalMain_web.m跑出来的结果分别代表什么意思,那个结果是论文中提到的结果呢,麻烦您指教一下可以吗,谢谢

SHERRYMINE commented 1 year ago

你可以再跑一遍,注意看那两个matlab文件在命令行跑出来的结果。 mean acc是指平均准确率,mean comp是指平均完整率,具体的评价指标在论文中有写,表示使用mvsnet模型训练后融合生成的点云和ground truth点云的之间的差别。

直观上的理解是这样的,0.4866 mean acc可以理解成:在生成的点云中,平均有48.66%的点深度和gt是一样的(在可接受的范围内);0.3943 mean comp可以理解成:生成的点云数量平均占gt的39.43%

SHERRYMINE commented 1 year ago

去看Read me的Results on DTU

SHERRYMINE commented 1 year ago

Deep Learning for Multi-view Stereo via Plane Sweep A Survey 我想你看完这一篇论文就都会了

lgc1030 commented 1 year ago

Deep Learning for Multi-view Stereo via Plane Sweep A Survey 我想你看完这一篇论文就都会了

好的

lgc1030 commented 1 year ago

Deep Learning for Multi-view Stereo via Plane Sweep A Survey 我想你看完这一篇论文就都会了

您好,请问提供的matlab代码中,哪一个matlab文件跑出来的结果是论文当中的acc和comp呀

SHERRYMINE commented 1 year ago

comstat的那个-------- 原始邮件 --------发件人: lihaha @.>日期: 2023年5月20日周六 21:24收件人: xy-guo/MVSNet_pytorch @.>抄送: SHERRYMINE @.>, Comment @.>主 题: Re: [xy-guo/MVSNet_pytorch] 代码中的matlab问题 (Issue #54)

Deep Learning for Multi-view Stereo via Plane Sweep A Survey 我想你看完这一篇论文就都会了

您好,请问提供的matlab代码中,哪一个matlab文件跑出来的结果是论文当中的acc和comp呀

—Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you commented.Message ID: @.***>

ZooKangaroo commented 9 months ago

@SHERRYMINE 请问如果我需要定量评价自己数据集的重建效果,需要怎么操作?如果使用这里给的matlab代码的话,那么obsmask文件是什么?该怎样生成?谢谢!