Df2net 3D model deployment

Through unremitting efforts ,DF2NET Finally in my little MAC Deployed . There are many problems , Here's a record .

github link ：https://github.com/xiaoxingzeng/DF2Net/

1. Environmental requirements

because 0.4.0 Version of pytorch Installation error , Here, the version is modified to 0.4.1, Use pip Installation .

about Dlib and PIL, You can install all the latest versions .

2. Code changes

DF2NET Of demo.py The input pictures of file processing are all from img_list.txt File read , And careful observation can find img_list.txt The contents of the file and the files contained in the file directory do not match one by one , Some files are missing , Need to find new pictures to add img Directory or modification img_list.txt File to match .

、

3. Problems caused by version differences

stay mac os in ,macos I won't support it cuda, So here will be num_workers Change it to 0, Prevent thread conflicts , And will all cuda Change it to cpu Carry out operations .

meanwhile , The old version of torch.autograd.Variable The method of using has changed , In the new torch In the version , New methods are needed to use Variable function , Otherwise, user level exceptions will appear .

Running results

Input picture ：

Using online matlab View and preview the model depth map .

Get the depth map .

Generate the final 3D model by running the detail drawing .