Jianzhu Guo comments

Results 126 comments of


                                            Jianzhu Guo

sh ./build.sh help me

You can run the command scripts in build.sh on Windows terminal and for some details, refer to https://github.com/cleardusk/3DDFA_V2/issues/12#issuecomment-697479173.

A question about details of the paper

I see. Thanks for your detailed reply!

question about the result with the parameters set as the code

Greatly thanks for releasing the code! Does that mean the final accuracy is determined by the pre-trained model? @yaoyao-liu I also pre-trained the model and get about **54.95%** on 5-way...

question about the result with the parameters set as the code

Thanks for your reply, I will try it.

question about the result with the parameters set as the code

I try [download_weight.sh](https://github.com/Sha-Lab/FEAT/blob/master/download_weight.sh) in [FEAT](https://github.com/Sha-Lab/FEAT) but fail to download it. 1) My terminal can not use VPN. 2) I copy the url to chrome (used with VPN) but it shows...

question about the result with the parameters set as the code

In my case, I use the uploaded pre-trained model but the accuracy is similar to training from scratch in the pre-train phase (very worse). Maybe there is the version conflict...

question about the result with the parameters set as the code

That really depends on the luck or ensembling for the pre-train phase may be a good choice.

question about the result with the parameters set as the code

Just an additional comment, I use an own pre-trained model trained longer and the 1/5 shot results are: ``` Val Best Epoch 43, Acc 0.6157, Test Acc 0.5797 Test Acc...

Source code and data

Sorry for the delay, the server storing the data and code breaks down, and it will be fixed in the next few days. Besides, my energy is further compressed due...

Downloading

Except Baidu Yun, I have no enough storage space on, e.g., google drive, onedrive.