No documentation for option to choose face detector
Allexxann opened this issue · comments
The GUI has 3 options for Face Detector and 3 options for Landmark Detector.
The command line parameters here list only a single model input with several different models, names sligtly off from the GUI.
These do not map onto each other.
How do I actually specify both detectors through command line?
Are you referring to "https://github.com/TadasBaltrusaitis/OpenFace/wiki/Command-line-arguments#documentation"?
Finally you might need to look into the implementation of the different samples/GUIs, like "https://github.com/TadasBaltrusaitis/OpenFace/blob/master/exe/FaceLandmarkImg/FaceLandmarkImg.cpp".
Here:
https://github.com/TadasBaltrusaitis/OpenFace/wiki/Command-line-arguments#common-parameters-for-all
Only one "-mloc" and it's all about landmark detection. GUI also has different options for face detection.
I don't know C/C++, so that gives me nothing.
Also, I like this GUI just fine. I just need to replicate all the parameters through command line - first I use the GUI to find the most useful combination of settings, then I batch process all the videos in my dataset with it.
Can you maybe rephrase your question @Allexxann ?
Which GUI executable are you talking about? We might want to check the code together to find all supported command-line arguments.
Are you asking about undocumented/outdated command-line arguments or do you want options from within the GUI only to be exposed as command-line parameters?
I simply use OpenFaceOffline.exe as a GUI executable, but from what I understand it simply passes the same calls internally.
Since I work with single-face videos, I'll need FeatureExtraction for command line input, I guess.
Hm, I honestly thought these options were already exposed, just not documented on that page.
Okay, more specific.
I want to know command line arguments
- to specify Face Detector - OpenCV(Haar)/Dlib(HOG-SVM)/OpenFace
- to specify Landmark Detector - CLM/CLNF/CE-CLM - I guess that's what -mloc is, but better safe than sorry
So that I could use through command line the mode I chose in GUI.
Everything else is mostly covered and is matched between documentation and GUI.
From what I can see in the files under "https://github.com/TadasBaltrusaitis/OpenFace/tree/master/gui/OpenFaceOffline" it looks like this GUI executable doesn't handle any command-line options, when I interpret the C# code correctly...
Everything seems hard-wired and only changable at runtime via the GUI...
OpenFaceOffline lacks console input, as it should, since it's a GUI integrator.
And FeatureExtraction lacks the GUI and operates through command line or being called by OpenFaceOffline.
That much I could tell you from the start.
Also I already checked FeatureExtraction (just a full-text regex search), and couldn't find any of the parameters listed in documentation, meaning that they're either handled elsewhere or encoded in a non-obvious way (I still know nothing about C++ and how it handles input).
You could check OpenFaceOffline for command line outputs/external calls, or variables set by choosing the options in GUI.
FeatureExtraction forwards the command-line arguments to various sub-modules, spread all over the repo:
LandmarkDetector::FaceModelParameters det_parameters(arguments);
FaceAnalysis::FaceAnalyserParameters face_analysis_params(arguments);
Utilities::Visualizer visualizer(arguments);
Utilities::RecorderOpenFaceParameters recording_params(arguments,... ...
Utilities::RecorderOpenFace open_face_rec(sequence_reader.name, recording_params, arguments);
The GUI OpenFaceOffline, however, supports all sub-modules, which then can be used via GUI buttons.
Hmm, still not sure I get your intention...
After finetuning the results with one of the GUI-tools you want to use one of the other executable, e.g. "https://github.com/TadasBaltrusaitis/OpenFace/tree/master/exe/FeatureExtraction", to do batching.
But you are missing command-line options for e.g. "FeatureExtraction.exe" or don't know their exact name?
Basically your last statement.
I'm sure the option for specifying face detector is already here, but the name is absent in the doc.
I just need the name of the parameter and the possible values for it.
It would be nice to confirm that "-mloc" is the option for landmark detector, though, just to be sure.
The latest version of "https://github.com/TadasBaltrusaitis/OpenFace/tree/master/gui/OpenFaceOffline" on the master branch doesn't consider any command line arguments, as far as I can see.
"After finetuning the results with one of the GUI-tools you want to use one of the other executable, e.g. "https://github.com/TadasBaltrusaitis/OpenFace/tree/master/exe/FeatureExtraction", to do batching.
But you are missing command-line options for e.g. "FeatureExtraction.exe" or don't know their exact name?"
I need arguments to pass to FeatureExtraction, which it passes further, but I don't know their exact name.
OpenFaceOffline is generating these arguments, not receiving them. At least, that's what it looks like from the user point of view.
Were you able to achieve what you were looking for, have you added your needed command line arguments and use them instead of the hard-coded parameters in this sample?
I tried to screw around with the source code, understood I have no hope before my deadline in making sense of C/C++, which I never studied before, then just used the default settings and hoped for the best.
First, the answer to my original question is not documented, but I hired a real c++ coder for a little detective work.
Supplying -fdloc "classifiers/haarcascade_frontalface_alt.xml" to FeatureExtraction forces the program to use Haar. There are several different xml's here, so you can choose whichever suits you.
Supplying -wild makes it use MTCNN.
If -wild is used, but it cannot find the file "model/mtcnn_detector/MTCNN_detector.txt", it will default to HOG instead.
Second, bmarkus, you're a fucking moron, and your "help" was less than useless. It is a big temptation to make yourself look clever without actually turning your brain on, but consider hanging yourself next time instead.
