yfeng95 / PRNet

Joint 3D Face Reconstruction and Dense Alignment with Position Map Regression Network (ECCV 2018)
http://openaccess.thecvf.com/content_ECCV_2018/papers/Yao_Feng_Joint_3D_Face_ECCV_2018_paper.pdf
MIT License
4.95k stars 947 forks source link

Training code #105

Open sduscx opened 5 years ago

sduscx commented 5 years ago

Hi,as a newer to this, I failed to achieve the training code, could you please provide training code to me? I hope I can get your help, thank you very much. My email: m17864154809@163.com

doujl commented 5 years ago

Hi.I also need to train this network with a new training set. Don't know if you are successful?If you succeed, can you tell me how to do it?My email: liangdjsha@163.com,thank you!

LucienXian commented 5 years ago

@zhuixunforever Hi, I want to know the "--train_data_file“,how to generate it in the repo https://github.com/jnulzl/PRNet? Can you give some tips?

qiannvMao commented 5 years ago

@zhuixunforever @WWWJL 您好,能否将您的训练代码发我一份,我刚学习深度学习,复现训练代码出现问题,非常感谢。邮箱:mqn20170910@163.com

yexiaoxi01 commented 5 years ago

https://github.com/YadiraF/PRNet 就是这里面的效果图,嗯嗯。我有用作者提供的模型测试,生成了obj,用meshlab打开是3D人脸。但是像链接里发的那个视频里的效果是如何实现的? ------------------ 原始邮件 ------------------ 发件人: "zhuixunforever"notifications@github.com; 发送时间: 2019年7月9日(星期二) 中午11:36 收件人: "YadiraF/PRNet"PRNet@noreply.github.com; 抄送: "吴佳玲"1324244977@qq.com;"Mention"mention@noreply.github.com; 主题: Re: [YadiraF/PRNet] Training code (#105) 我其实没有看到你发的效果图。然后作者的代码里面有写3d点到obj里面,那个函数之前的函数就是把uv_map转化成3D点 — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.

你好,我想问问你有没有用自己训练的模型生成.obj. 我用meshlab打开的face shape 和 纹理贴图不匹配。不知道为什么。还有你对数据集里面的.mat文件了解吗, 他如何和BFM模型生成uv_posmap的,和3DMM模型是什么关系。虽然可以通过作者提供的代码生成uv图,但是原理很模糊。

yexiaoxi01 commented 5 years ago

@zhuixunforever 非常感谢你的回复,我会试着用这个来训练模型。还有,你对mat文件里面的信息和bfm模型如何应用了解吗。我一直对生成uv图的这个过程比较困惑。这个3dMM模型和bfm模型是怎么用的。再次感谢你的帮助-

yexiaoxi01 commented 5 years ago

@zhuixunforever 谢谢解惑啦,我先尝试进行训练,有问题再问你。谢谢谢谢

yexiaoxi01 commented 5 years ago

@zhuixunforever 你好,我是yexiaoxi01,在忙吗?之前请教你的那个问题,你推荐https://github.com/shen1994/DeepFaceRestruction 这个代码训练,我发现训练所有图像,很慢,大概需要近二十多天,还有就是,我如何通过训练好的模来设生成。obj文件,就像原作者那个run_basic.py. 我尝试好多修改,但是模型加载不出来,一直有问题。

vitahsu commented 4 years ago

您好! 請問paper內部提到的 "we use the parameterized UV coordinates which computes a Tutte embedding with conformal Laplacian weight and then maps the mesh boundary to a square" 有沒有詳細的code描述BFM_UV.mat是怎麼生成的? 謝謝!

zhao181 commented 4 years ago

您好! 請問paper內部提到的 "we use the parameterized UV coordinates which computes a Tutte embedding with conformal Laplacian weight and then maps the mesh boundary to a square" 有沒有詳細的code描述BFM_UV.mat是怎麼生成的? 謝謝!

您好同学! 请问你有没有找到详细资料介绍BFM_UV.mat生成的?

vitahsu commented 4 years ago

@zhao181 在3D morphable models as spatial transformer networks這篇paper的section2.3 output grid是此篇paper引用的原出處,可以看一下! 不過看過去也還是非常模糊啊!!!!! (可能背景知識還不足) 若有問題歡迎一起討論! 另外在此連結也有一些討論唷! 謝謝

zhao181 commented 4 years ago

谢谢 好的 On 11/21/2019 11:50,vitahsunotifications@github.com wrote:

@zhao181 在3D morphable models as spatial transformer networks這篇paper的section2.3 output grid是此篇paper引用的原出處,可以看一下! 不過看過去也還是非常模糊啊!!!!! (可能背景知識還不足) 若有問題歡迎一起討論! 另外在此連結也有一些討論唷! 謝謝

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.

vitahsu commented 4 years ago

@WWWJL 您好! 我最近使用tensorflow在復現paper,但是復現的磕磕拌拌的,想請問您是否願意分享您的訓練碼? 我的信箱為bantiauo@gmail.com,謝謝您! 另外若不方便分享的話想請教您,訓練300WLP這包6W多張Image在GPU上training的時間大概多久?謝謝!

cuixin1992 commented 3 years ago

@vitahsu @zhao181 您好!请问您们现在明白如何生成BFM_UV.mat了吗?