Generation AU Problem

[shliang0603]

First，I use the below command line to generation AU
E:\MagicProject\code\OpenFace_2.0.4_win_x64>FeatureExtraction.exe -f E:\Dataset_processing\20180821_GANimation_CK+\imgs\S005_001_00000001.png -out_dir E:\out

I generate CSV file of AU is from 679 column to 696，like below

Finally, I use this to train GANimation. But my train result is not expectation. I suspect that I generation is false.

Today ,I use FeatureExtraction.exe to generate AU ,but I choice different parameter，the command line and generation CSV of Au in below

1. generation AU from column 6 to column 25
   E:\MagicProject\code\OpenFace_2.0.4_win_x64>FeatureExtraction.exe -f E:\Dataset_processing\20180821_GANimation_CK+\imgs\S005_001_00000001.png -aus -out_dir E:\

E:\MagicProject\code\OpenFace_2.0.4_win_x64>FeatureExtraction.exe -f E:\Dataset_processing\20180821_GANimation_CK+\imgs\S005_001_00000001.png -aus -au_static -out_dir E:\out

But ,today I use 1 and 2 openface command line to generate AU ,the result is different.
(I used the same picture for the AU generation test)

Can you tell me how to use the right openface command parameter to generate CSV of AU.

[XiangyuWu]

I think it is FaceLandmarkImg rather than FeatureExtraction that you should use.

[xiaoiker]

@XiangyuWu You are right. BTW, is there something wrong with the content[2:19] in prepare_au_annotations.py? why not content[1:18]?

[XiangyuWu]

@xiaoiker I don't know how does your question come. In my csv file, content[0] is face, content[1] is confidence, and form content[2] to content[18] is exactly the value of action unit.

[xiaoiker]

@XiangyuWu Sorry, my fault. Thanks very much for pointing out.

[shliang0603]

@XiangyuWu Hi，Xiangyu Wu,Can you send me your the command line of generation AU(action unit).
In my file, generation AU is between 679 column and 696 column

[XiangyuWu]

@shliang0603 adding -aus to your command line. For linux platform, my command line is:
./FaceLandmarkImg -fdir path_to_images -out_dir path_for_generated_files -aus

[shliang0603]

@XiangyuWu Thanks for your answering, I will try again

[ilovecv]

Hi Xiangyu, did you find the detected AU results from OpenFace are different from the labels provided in aus_openface.pkl in the sample dataset?

For example, for N_0000000356_00190.jpg, the labels provided by aus_openface.pkl are:
[2.86, 2.27, 1.45, 1.1 , 0. , 0.65, 0.05, 0. , 0.75, 1.65, 0.6 , 0. , 1.86, 0. , 0.62, 0.25, 0. ],
while the detected results from OpenFace are:
[0.48, 0. , 0. , 1.47, 0.23, 0.06, 0. , 0.72, 0.48, 1.36, 0.3 , 0. , 1.76, 0. , 0.44, 0.78, 0. ].

I used the command: ./bin/FaceLandmarkImg -fdir GANimation/sample_dataset/imgs -out_dir GANimation/sample_dataset/features -aus

Thank you very much! @XiangyuWu

[XiangyuWu]

@ilovecv I didn't pay attention to it, and I tested it just now, the same problem happened to me. I don't know why such differences exist.

[shliang0603]

@ilovecv Unfortunately, it is really difficult to ensure the same exact numerical results when using different compilers (e.g. see https://stackoverflow.com/questions/16395615/is-there-any-way-to-make-sure-the-floating-point-arithmetic-result-the-same-in-b), the differences will come from compilers optimizing code differently and possibly different versions of libraries used with different optimization settings (e.g. OpenCV)

[albertpumarola]

Closed for inactivity