-
Notifications
You must be signed in to change notification settings - Fork 1.8k
Command line arguments
One way of interacting with OpenFace is through command line arguments (e.g. Command Prompt or PowerShell in Windows and xterm on Unix systems). For this you need to first compile the OpenFace code or download the executables (for Windows).
FeatureExtraction executable is used for sequence analysis that contain a single face, FaceLandmarkVidMulti is intended for sequence analysis that contain multiple faces, and FaceLandmarkImg executable is for individual image analysis (can either contain one or more faces). Some common example uses are outlined below. For Ubuntu or Mac OS X omit the .exe. For the commands to work the current directory has to be in the same directory as the executables, or you need to specify the full path to the executable.
If you want to extract OpenFace features (by features we refer to all the features extracted by OpenFace: facial landmarks, head pose, eye gaze, facial action units, similarity aligned faces, and HOG) from a video file in location C:\my videos\video.avi, assuming that only one person appears in that video file, execute the following command on the command line:
FeatureExtraction.exe -f "C:\my videos\video.avi"
If there are multiple people in the video use:
FaceLandmarkVidMulti.exe -f "C:\my videos\video.avi"
This will create a processed directory in the same directory (working directory) as FeatureExtraction or FaceLandmarkVidMulti that will contain the processed features (more details).
If you want to extract OpenFace features from multiple video files, e.g. C:\my videos\video1.avi, C:\my videos\video2.avi, C:\my videos\video3.avi, execute the following command on the command line:
FeatureExtraction.exe -f "C:\my videos\video1.avi" -f "C:\my videos\video2.avi" -f "C:\my videos\video3.avi"
or
FaceLandmarkVidMulti.exe -f "C:\my videos\video1.avi" -f "C:\my videos\video2.avi" -f "C:\my videos\video3.avi"
If you only want to extract head pose from the video file in C:\my videos\video1.avi, execute the following command on the command line:
FeatureExtraction.exe -f "C:\my videos\video1.avi" -pose
If instead of a video file you have a sequence of images in a directory with each image representing a frame (e.g. 001.jpg, 002.jpg, 003.jpg ... 200.jpg in directory "C:\my videos\sequence1"), you can extract features from that sequence:
FeatureExtraction.exe -fdir "C:\my videos\sequence1"
or
FaceLandmarkVidMulti.exe -fdir "C:\my videos\sequence1"
If you want to extract OpenFace features from an image file in location C:\my images\img.jpg (can contain multiple people), execute the following command on the command line:
FaceLandmarkImg.exe -f "C:\my images\img.jpg"
This will create a processed directory in the same directory (working directory) as FaceLandmarkImg that will contain the processed features (more details ).
If you want to extract OpenFace features from multiple image files, e.g. C:\my images\img1.jpg, C:\my images\img2.jpg, C:\my images\img3.jpg, execute the following command on the command line:
FaceLandmarkImg.exe -f "C:\my images\img1.jpg" -f "C:\my images\img2.jpg" -f "C:\my images\img3.jpg"
Alternatively, if you want to process all the images in the directory C:\my images, you can use:
FaceLandmarkImg.exe -fdir "C:\my images"
This will process all of the images (.jpg, jpeg, .bmp, .png files) in the directory.
NOTE, the difference between using -fdir flag between FeatureExtraction and FaceLandmarkImg is that FeatureExtraction assumes a sequence of images coming from a video that contains only one face, while FaceLandmarkImg assumes a random collection of images with no relationship between them, that can contain multiple faces.
Another good way to see how the command line arguments work is through looking at ./matlab_runners/Demos/ which illustrates how each of the models is called.
Single image analysis
-f <filename> the image file being input, can have multiple -f flags
-out_dir <directory> name of the output directory, where processed features will be places (i.e. CSV file for landmarks, gaze, and aus, HOG feature file, image with detected landmarks and a meta file)
-root <dir> the root directory so -f can be specified relative to it
-inroot <dir> the input root directory so -f can be specified relative to it
Batch image analysis
-fdir <directory> - runs landmark detection on all images (.jpg, .jpeg, .png, and .bmp) in a directory
-bboxdir optional directory that contains .txt files (image_name.txt) with bounding box (min_x min_y max_x max_y), it will use those for initialisation instead of an internal face detector
-out_dir <directory> name of the output directory, where processed features will be places (i.e. CSV file for landmarks, gaze, and aus, HOG feature file, image with detected landmarks and a meta file)
For more details on output format see here
Input parameters
-f <filename> the video file being input, can specify multiple -f
-fdir <directory> run the feature extraction on every image (.jpg, .jpeg, .png, and .bmp) in a directory (the output will be stored in individual files for the whole directory)
-device <device id> the device id of a webcam to perform feature extraction from a live feed
-cam_width the input webcam witch resolution (default 640), only valid if device is set
-cam_height the input webcam height resolution (default 480), only valid if device is set
-root <dir> the root for input
-inroot <dir> the root for input
-au_static if this flag is specified the AU prediction will be performed as if on static images rather than videos, see here for a more detailed explanation, this is the default behavior for FaceLandmarkVidMulti
Parameters for output (for more details on output format see here)
-out_dir <dir> the root directory relevant to which the output files are created
-of <filename> the filename of the output (if not specified will use original filename, the name of the directory in case of -fdir and webcam + timestamp in case of a webcam)
-oc <FOURCC_CODE> the codec of the output video file (list of FOURCC codes can be found here - https://www.fourcc.org/codecs.php)
Additional parameters for output (applicable for both FaceLandmarkImg and FeatureExtraction)
-verbose visualise the processing steps live: tracked face with gaze, action units, similarity aligned face, and HOG features (not visualized by default), this flag turns all of them on, below flags allow for more fine-grained control. Visualizing these outputs will reduce processing speeds, potentially by a significant amount.
-vis-track visualise the tracked face
-vis-hog visualise the HOG features
-vis-align visualise similarity aligned faces
-vis-aus visualise Action Units
-simscale <float> scale of the face for similarity alignment (default 0.7)
-simsize <int> width and height of image in pixels when similarity aligned (default 112)
-format_aligned <format> output image format for aligned faces (e.g. png or jpg), any format supported by OpenCV
-format_vis_image <format> output image format for visualized images (e.g. png or jpg), any format supported by OpenCV. Only applicable to FaceLandmarkImg
-nomask forces the aligned face output images to not be masked out
-g output images should be grayscale (for saving space)
By default the executable will output all features (tracked videos, HOG files, similarity aligned images and a .csv file with landmarks, action units and gaze). You might not always want to extract all the output features, you can specify the desired output using the following flags:
-2Dfp output 2D landmarks in pixels
-3Dfp output 3D landmarks in milimeters
-pdmparams output rigid and non-rigid shape parameters
-pose output head pose (location and rotation)
-aus output the Facial Action Units
-gaze output gaze and related features (2D and 3D locations of eye landmarks)
-hogalign output extracted HOG feaure file
-simalign output similarity aligned images of the tracked faces
-nobadaligned if outputting similarity aligned images, do not output from frames where detection failed or is unreliable (thus saving some disk space)
-tracked output video with detected landmarks
Parameters for input
-f <filename> the video file being input, can specify multiple files by having multiple -f flags
-device <device_num> the webcam from which to read images (default 0)
-cam_width the input webcam witch resolution (default 640)
-cam_height the input webcam height resolution (default 480)
-inroot <directory> the root of input so -f can be specified relative to it
Model to use parameters
-mloc <the location of landmark detection models>
Options for this:
-
"model/main_ceclm_general.txt"(default) - trained on Multi-PIE of varying pose and illumination and In-the-wild data (300W and Menpo). Most accurate, but the slowest and most memory heavy method. Suitable for most conditions. -
"model/main_clnf_general.txt"- trained on Multi-PIE of varying pose and illumination and In-the-wild data. Works well in a lot of conditions, but fails for profile faces and very large expressions -
"model/main_clnf_wild.txt"- as above, but only trained on 300W datasets (Helen and LFPW) -
"model/main_clnf_multi_pie.txt"- trained on CMU Multi-PIE dataset, works well in well lit environments -
"model/main_clm_general.txt"- the least accurate. but the fastest model
Model parameters
-wild flag specifies when the images are more difficult, model considers extended search regions
-multi_view <0/1> should multi-view initialisation be used (more robust, but slower), off by default
Optional camera parameters for proper head pose and eye gaze computation
-fx <focal length in x>
-fy <focal length in y>
-cx <optical centre in x>
-cy <optical centre in y>