def undistort_points(config, dataframe, camera_pair):
cfg_3d = auxiliaryfunctions.read_config(config)
path_camera_matrix = auxiliaryfunctions_3d.Foldernames3Dproject(cfg_3d)[2]
"""
path_undistort = destfolder
filename_cam1 = Path(dataframe[0]).stem
filename_cam2 = Path(dataframe[1]).stem
#currently no interm. saving of this due to high speed.
# check if the undistorted files are already present
if os.path.exists(os.path.join(path_undistort,filename_cam1 + '_undistort.h5')) and os.path.exists(os.path.join(path_undistort,filename_cam2 + '_undistort.h5')):
print("The undistorted files are already present at %s" % os.path.join(path_undistort,filename_cam1))
dataFrame_cam1_undistort = pd.read_hdf(os.path.join(path_undistort,filename_cam1 + '_undistort.h5'))
dataFrame_cam2_undistort = pd.read_hdf(os.path.join(path_undistort,filename_cam2 + '_undistort.h5'))
else:
"""
# Create an empty dataFrame to store the undistorted 2d coordinates and likelihood
dataframe_cam1 = pd.read_hdf(dataframe[0])
dataframe_cam2 = pd.read_hdf(dataframe[1])
path_stereo_file = os.path.join(path_camera_matrix, "stereo_params.pickle")
stereo_file = auxiliaryfunctions.read_pickle(path_stereo_file)
dataFrame_cam1_undistort, dataFrame_cam2_undistort = _undistort_views(
[(dataframe_cam1, dataframe_cam2)], stereo_file,
)[0]
return (
dataFrame_cam1_undistort,
dataFrame_cam2_undistort,
stereo_file[camera_pair],
path_stereo_file,
)
def create_video_from_pickled_tracks(
video, pickle_file, destfolder="", output_name="", pcutoff=0.6
):
if not destfolder:
destfolder = os.path.splitext(video)[0]
if not output_name:
video_name, ext = os.path.splitext(os.path.split(video)[1])
output_name = video_name + "DLClabeled" + ext
tracks = auxiliaryfunctions.read_pickle(pickle_file)
_create_video_from_tracks(video, tracks, destfolder, output_name, pcutoff)
def undistort_points(config, dataframe, camera_pair):
cfg_3d = auxiliaryfunctions.read_config(config)
(
img_path,
path_corners,
path_camera_matrix,
path_undistort,
) = auxiliaryfunctions_3d.Foldernames3Dproject(cfg_3d)
"""
path_undistort = destfolder
filename_cam1 = Path(dataframe[0]).stem
filename_cam2 = Path(dataframe[1]).stem
#currently no interm. saving of this due to high speed.
# check if the undistorted files are already present
if os.path.exists(os.path.join(path_undistort,filename_cam1 + '_undistort.h5')) and os.path.exists(os.path.join(path_undistort,filename_cam2 + '_undistort.h5')):
print("The undistorted files are already present at %s" % os.path.join(path_undistort,filename_cam1))
dataFrame_cam1_undistort = pd.read_hdf(os.path.join(path_undistort,filename_cam1 + '_undistort.h5'))
dataFrame_cam2_undistort = pd.read_hdf(os.path.join(path_undistort,filename_cam2 + '_undistort.h5'))
else:
"""
if True:
# Create an empty dataFrame to store the undistorted 2d coordinates and likelihood
dataframe_cam1 = pd.read_hdf(dataframe[0])
dataframe_cam2 = pd.read_hdf(dataframe[1])
scorer_cam1 = dataframe_cam1.columns.get_level_values(0)[0]
scorer_cam2 = dataframe_cam2.columns.get_level_values(0)[0]
path_stereo_file = os.path.join(path_camera_matrix,
"stereo_params.pickle")
stereo_file = auxiliaryfunctions.read_pickle(path_stereo_file)
mtx_l = stereo_file[camera_pair]["cameraMatrix1"]
dist_l = stereo_file[camera_pair]["distCoeffs1"]
mtx_r = stereo_file[camera_pair]["cameraMatrix2"]
dist_r = stereo_file[camera_pair]["distCoeffs2"]
R1 = stereo_file[camera_pair]["R1"]
P1 = stereo_file[camera_pair]["P1"]
R2 = stereo_file[camera_pair]["R2"]
P2 = stereo_file[camera_pair]["P2"]
# Create an empty dataFrame to store the undistorted 2d coordinates and likelihood
(
dataFrame_cam1_undistort,
scorer_cam1,
bodyparts,
) = auxiliaryfunctions_3d.create_empty_df(dataframe_cam1,
scorer_cam1,
flag="2d")
(
dataFrame_cam2_undistort,
scorer_cam2,
bodyparts,
) = auxiliaryfunctions_3d.create_empty_df(dataframe_cam2,
scorer_cam2,
flag="2d")
for bpindex, bp in tqdm(enumerate(bodyparts)):
points_cam1 = dataframe_cam1.xs(bp, level="bodyparts",
axis=1).values[:, :2]
points_cam1_remapped = cv2.undistortPoints(
src=points_cam1.astype(np.float32),
cameraMatrix=mtx_l,
distCoeffs=dist_l,
P=P1,
R=R1,
)
dataFrame_cam1_undistort.loc(
axis=1)[scorer_cam1, bp,
["x", "y"]] = points_cam1_remapped.squeeze()
dataFrame_cam1_undistort.loc(
axis=1)[scorer_cam1, bp, "likelihood"] = dataframe_cam1.xs(
[bp, "likelihood"], level=["bodyparts",
"coords"], axis=1).values
# Undistorting the points from cam2 camera
points_cam2 = dataframe_cam2.xs(bp, level="bodyparts",
axis=1).values[:, :2]
points_cam2_remapped = cv2.undistortPoints(
src=points_cam2.astype(np.float32),
cameraMatrix=mtx_r,
distCoeffs=dist_r,
P=P2,
R=R2,
)
dataFrame_cam2_undistort.loc(
axis=1)[scorer_cam2, bp,
["x", "y"]] = points_cam2_remapped.squeeze()
dataFrame_cam2_undistort.loc(
axis=1)[scorer_cam2, bp, "likelihood"] = dataframe_cam2.xs(
[bp, "likelihood"], level=["bodyparts",
"coords"], axis=1).values
# Save the undistorted files
dataFrame_cam1_undistort.sort_index(inplace=True)
dataFrame_cam2_undistort.sort_index(inplace=True)
return (
dataFrame_cam1_undistort,
dataFrame_cam2_undistort,
stereo_file[camera_pair],
path_stereo_file,
)
def triangulate(
config,
video_path,
videotype="avi",
filterpredictions=True,
filtertype="median",
gputouse=None,
destfolder=None,
save_as_csv=False,
):
"""
This function triangulates the detected DLC-keypoints from the two camera views
using the camera matrices (derived from calibration) to calculate 3D predictions.
Parameters
----------
config : string
Full path of the config.yaml file as a string.
video_path : string/list of list
Full path of the directory where videos are saved. If the user wants to analyze
only a pair of videos, the user needs to pass them as a list of list of videos,
i.e. [['video1-camera-1.avi','video1-camera-2.avi']]
videotype: string, optional
Checks for the extension of the video in case the input to the video is a directory.\n
Only videos with this extension are analyzed. The default is ``.avi``
filterpredictions: Bool, optional
Filter the predictions with filter specified by "filtertype". If specified it
should be either ``True`` or ``False``.
filtertype: string
Select which filter, 'arima' or 'median' filter (currently supported).
gputouse: int, optional. Natural number indicating the number of your GPU (see number in nvidia-smi).
If you do not have a GPU put None.
See: https://nvidia.custhelp.com/app/answers/detail/a_id/3751/~/useful-nvidia-smi-queries
destfolder: string, optional
Specifies the destination folder for analysis data (default is the path of the video)
save_as_csv: bool, optional
Saves the predictions in a .csv file. The default is ``False``
Example
-------
Linux/MacOS
To analyze all the videos in the directory:
>>> deeplabcut.triangulate(config,'/data/project1/videos/')
To analyze only a few pairs of videos:
>>> deeplabcut.triangulate(config,[['/data/project1/videos/video1-camera-1.avi','/data/project1/videos/video1-camera-2.avi'],['/data/project1/videos/video2-camera-1.avi','/data/project1/videos/video2-camera-2.avi']])
Windows
To analyze all the videos in the directory:
>>> deeplabcut.triangulate(config,'C:\\yourusername\\rig-95\\Videos')
To analyze only a few pair of videos:
>>> deeplabcut.triangulate(config,[['C:\\yourusername\\rig-95\\Videos\\video1-camera-1.avi','C:\\yourusername\\rig-95\\Videos\\video1-camera-2.avi'],['C:\\yourusername\\rig-95\\Videos\\video2-camera-1.avi','C:\\yourusername\\rig-95\\Videos\\video2-camera-2.avi']])
"""
from deeplabcut.pose_estimation_tensorflow import predict_videos
from deeplabcut.post_processing import filtering
cfg_3d = auxiliaryfunctions.read_config(config)
cam_names = cfg_3d["camera_names"]
pcutoff = cfg_3d["pcutoff"]
scorer_3d = cfg_3d["scorername_3d"]
snapshots = {}
for cam in cam_names:
snapshots[cam] = cfg_3d[str("config_file_" + cam)]
# Check if the config file exists
if not os.path.exists(snapshots[cam]):
raise Exception(
str("It seems the file specified in the variable config_file_"
+ str(cam)) +
" does not exist. Please edit the config file with correct file path and retry."
)
# flag to check if the video_path variable is a string or a list of list
flag = False # assumes that video path is a list
if isinstance(video_path, str) == True:
flag = True
video_list = auxiliaryfunctions_3d.get_camerawise_videos(
video_path, cam_names, videotype=videotype)
else:
video_list = video_path
if video_list == []:
print("No videos found in the specified video path.", video_path)
print(
"Please make sure that the video names are specified with correct camera names as entered in the config file or"
)
print(
"perhaps the videotype is distinct from the videos in the path, I was looking for:",
videotype,
)
print("List of pairs:", video_list)
scorer_name = {}
run_triangulate = False
for i in range(len(video_list)):
dataname = []
for j in range(len(video_list[i])): # looping over cameras
if cam_names[j] in video_list[i][j]:
print("Analyzing video %s using %s" %
(video_list[i][j], str("config_file_" + cam_names[j])))
config_2d = snapshots[cam_names[j]]
cfg = auxiliaryfunctions.read_config(config_2d)
shuffle = cfg_3d[str("shuffle_" + cam_names[j])]
trainingsetindex = cfg_3d[str("trainingsetindex_" +
cam_names[j])]
trainFraction = cfg["TrainingFraction"][trainingsetindex]
if flag == True:
video = os.path.join(video_path, video_list[i][j])
else:
video_path = str(Path(video_list[i][j]).parents[0])
video = os.path.join(video_path, video_list[i][j])
if destfolder is None:
destfolder = str(Path(video).parents[0])
vname = Path(video).stem
prefix = str(vname).split(cam_names[j])[0]
suffix = str(vname).split(cam_names[j])[-1]
if prefix == "":
pass
elif prefix[-1] == "_" or prefix[-1] == "-":
prefix = prefix[:-1]
if suffix == "":
pass
elif suffix[0] == "_" or suffix[0] == "-":
suffix = suffix[1:]
if prefix == "":
output_file = os.path.join(destfolder, suffix)
else:
if suffix == "":
output_file = os.path.join(destfolder, prefix)
else:
output_file = os.path.join(destfolder,
prefix + "_" + suffix)
output_filename = os.path.join(
output_file + "_" + scorer_3d
) # Check if the videos are already analyzed for 3d
if os.path.isfile(output_filename + ".h5"):
if save_as_csv is True and not os.path.exists(
output_filename + ".csv"):
# In case user adds save_as_csv is True after triangulating
pd.read_hdf(output_filename + ".h5").to_csv(
str(output_filename + ".csv"))
print(
"Already analyzed...Checking the meta data for any change in the camera matrices and/or scorer names",
vname,
)
pickle_file = str(output_filename + "_meta.pickle")
metadata_ = auxiliaryfunctions_3d.LoadMetadata3d(
pickle_file)
(
img_path,
path_corners,
path_camera_matrix,
path_undistort,
) = auxiliaryfunctions_3d.Foldernames3Dproject(cfg_3d)
path_stereo_file = os.path.join(path_camera_matrix,
"stereo_params.pickle")
stereo_file = auxiliaryfunctions.read_pickle(
path_stereo_file)
cam_pair = str(cam_names[0] + "-" + cam_names[1])
if_video_analyzed = False # variable to keep track if the video was already analyzed
# Check for the camera matrix
for k in metadata_["stereo_matrix"].keys():
if np.all(metadata_["stereo_matrix"][k] ==
stereo_file[cam_pair][k]):
pass
else:
run_triangulate = True
# Check for scorer names in the pickle file of 3d output
DLCscorer, DLCscorerlegacy = auxiliaryfunctions.GetScorerName(
cfg,
shuffle,
trainFraction,
trainingsiterations="unknown")
if (metadata_["scorer_name"][cam_names[j]] == DLCscorer
): # TODO: CHECK FOR BOTH?
if_video_analyzed = True
elif metadata_["scorer_name"][
cam_names[j]] == DLCscorerlegacy:
if_video_analyzed = True
else:
if_video_analyzed = False
run_triangulate = True
if if_video_analyzed:
print("This file is already analyzed!")
dataname.append(
os.path.join(destfolder,
vname + DLCscorer + ".h5"))
scorer_name[cam_names[j]] = DLCscorer
else:
# Analyze video if score name is different
DLCscorer = predict_videos.analyze_videos(
config_2d,
[video],
videotype=videotype,
shuffle=shuffle,
trainingsetindex=trainingsetindex,
gputouse=gputouse,
destfolder=destfolder,
)
scorer_name[cam_names[j]] = DLCscorer
if_video_analyzed = False
run_triangulate = True
if filterpredictions:
filtering.filterpredictions(
config_2d,
[video],
videotype=videotype,
shuffle=shuffle,
trainingsetindex=trainingsetindex,
filtertype=filtertype,
destfolder=destfolder,
)
dataname.append(
os.path.join(destfolder,
vname + DLCscorer + ".h5"))
else: # need to do the whole jam.
DLCscorer = predict_videos.analyze_videos(
config_2d,
[video],
videotype=videotype,
shuffle=shuffle,
trainingsetindex=trainingsetindex,
gputouse=gputouse,
destfolder=destfolder,
)
scorer_name[cam_names[j]] = DLCscorer
run_triangulate = True
print(destfolder, vname, DLCscorer)
if filterpredictions:
filtering.filterpredictions(
config_2d,
[video],
videotype=videotype,
shuffle=shuffle,
trainingsetindex=trainingsetindex,
filtertype=filtertype,
destfolder=destfolder,
)
dataname.append(
os.path.join(destfolder,
vname + DLCscorer + ".h5"))
if run_triangulate:
# if len(dataname)>0:
# undistort points for this pair
print("Undistorting...")
(
dataFrame_camera1_undistort,
dataFrame_camera2_undistort,
stereomatrix,
path_stereo_file,
) = undistort_points(config, dataname,
str(cam_names[0] + "-" + cam_names[1]))
if len(dataFrame_camera1_undistort) != len(
dataFrame_camera2_undistort):
import warnings
warnings.warn(
"The number of frames do not match in the two videos. Please make sure that your videos have same number of frames and then retry! Excluding the extra frames from the longer video."
)
if len(dataFrame_camera1_undistort) > len(
dataFrame_camera2_undistort):
dataFrame_camera1_undistort = dataFrame_camera1_undistort[:len(
dataFrame_camera2_undistort)]
if len(dataFrame_camera2_undistort) > len(
dataFrame_camera1_undistort):
dataFrame_camera2_undistort = dataFrame_camera2_undistort[:len(
dataFrame_camera1_undistort)]
# raise Exception("The number of frames do not match in the two videos. Please make sure that your videos have same number of frames and then retry!")
X_final = []
triangulate = []
scorer_cam1 = dataFrame_camera1_undistort.columns.get_level_values(
0)[0]
scorer_cam2 = dataFrame_camera2_undistort.columns.get_level_values(
0)[0]
df_3d, scorer_3d, bodyparts = auxiliaryfunctions_3d.create_empty_df(
dataFrame_camera1_undistort, scorer_3d, flag="3d")
P1 = stereomatrix["P1"]
P2 = stereomatrix["P2"]
print("Computing the triangulation...")
for bpindex, bp in enumerate(bodyparts):
# Extract the indices of frames where the likelihood of a bodypart for both cameras are less than pvalue
likelihoods = np.array([
dataFrame_camera1_undistort[scorer_cam1][bp]
["likelihood"].values[:],
dataFrame_camera2_undistort[scorer_cam2][bp]
["likelihood"].values[:],
])
likelihoods = likelihoods.T
# Extract frames where likelihood for both the views is less than the pcutoff
low_likelihood_frames = np.any(likelihoods < pcutoff, axis=1)
# low_likelihood_frames = np.all(likelihoods < pcutoff, axis=1)
low_likelihood_frames = np.where(
low_likelihood_frames == True)[0]
points_cam1_undistort = np.array([
dataFrame_camera1_undistort[scorer_cam1][bp]
["x"].values[:],
dataFrame_camera1_undistort[scorer_cam1][bp]
["y"].values[:],
])
points_cam1_undistort = points_cam1_undistort.T
# For cam1 camera: Assign nans to x and y values of a bodypart where the likelihood for is less than pvalue
points_cam1_undistort[low_likelihood_frames] = np.nan, np.nan
points_cam1_undistort = np.expand_dims(points_cam1_undistort,
axis=1)
points_cam2_undistort = np.array([
dataFrame_camera2_undistort[scorer_cam2][bp]
["x"].values[:],
dataFrame_camera2_undistort[scorer_cam2][bp]
["y"].values[:],
])
points_cam2_undistort = points_cam2_undistort.T
# For cam2 camera: Assign nans to x and y values of a bodypart where the likelihood is less than pvalue
points_cam2_undistort[low_likelihood_frames] = np.nan, np.nan
points_cam2_undistort = np.expand_dims(points_cam2_undistort,
axis=1)
X_l = auxiliaryfunctions_3d.triangulatePoints(
P1, P2, points_cam1_undistort, points_cam2_undistort)
# ToDo: speed up func. below by saving in numpy.array
X_final.append(X_l)
triangulate.append(X_final)
triangulate = np.asanyarray(triangulate)
metadata = {}
metadata["stereo_matrix"] = stereomatrix
metadata["stereo_matrix_file"] = path_stereo_file
metadata["scorer_name"] = {
cam_names[0]: scorer_name[cam_names[0]],
cam_names[1]: scorer_name[cam_names[1]],
}
# Create an empty dataframe to store x,y,z of 3d data
for bpindex, bp in enumerate(bodyparts):
df_3d.iloc[:][scorer_3d, bp, "x"] = triangulate[0, bpindex,
0, :]
df_3d.iloc[:][scorer_3d, bp, "y"] = triangulate[0, bpindex,
1, :]
df_3d.iloc[:][scorer_3d, bp, "z"] = triangulate[0, bpindex,
2, :]
df_3d.to_hdf(
str(output_filename + ".h5"),
"df_with_missing",
format="table",
mode="w",
)
auxiliaryfunctions_3d.SaveMetadata3d(
str(output_filename + "_meta.pickle"), metadata)
if save_as_csv:
df_3d.to_csv(str(output_filename + ".csv"))
print("Triangulated data for video", video_list[i])
print("Results are saved under: ", destfolder)
# have to make the dest folder none so that it can be updated for a new pair of videos
if destfolder == str(Path(video).parents[0]):
destfolder = None
if len(video_list) > 0:
print("All videos were analyzed...")
print(
"Now you can create 3D video(s) using deeplabcut.create_labeled_video_3d"
)
def check_undistortion(config, cbrow=8, cbcol=6, plot=True):
"""
This function undistorts the calibration images based on the camera matrices and stores them in the project folder(defined in the config file)
to visually check if the camera matrices are correct.
Parameters
----------
config : string
Full path of the config.yaml file as a string.
cbrow : int
Int specifying the number of rows in the calibration image.
cbcol : int
Int specifying the number of columns in the calibration image.
plot : bool
If this is set to True, the results of undistortion are saved as plots. The default is ``True``; if provided it must be either ``True`` or ``False``.
Example
--------
Linux/MacOs/Windows
>>> deeplabcut.check_undistortion(config, cbrow = 8,cbcol = 6)
"""
# Read the config file
criteria = (cv2.TERM_CRITERIA_EPS + cv2.TERM_CRITERIA_MAX_ITER, 30, 0.001)
cfg_3d = auxiliaryfunctions.read_config(config)
(
img_path,
path_corners,
path_camera_matrix,
path_undistort,
) = auxiliaryfunctions_3d.Foldernames3Dproject(cfg_3d)
# colormap = plt.get_cmap(cfg_3d['colormap'])
markerSize = cfg_3d["dotsize"]
alphaValue = cfg_3d["alphaValue"]
markerType = cfg_3d["markerType"]
markerColor = cfg_3d["markerColor"]
cam_names = cfg_3d["camera_names"]
images = glob.glob(os.path.join(img_path, "*.jpg"))
# Sort the images
images.sort(key=lambda f: int("".join(filter(str.isdigit, f))))
"""
for fname in images:
for cam in cam_names:
if cam in fname:
filename = Path(fname).stem
ext = Path(fname).suffix
img = cv2.imread(fname)
gray = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)
"""
camera_pair = [[cam_names[0], cam_names[1]]]
stereo_params = auxiliaryfunctions.read_pickle(
os.path.join(path_camera_matrix, "stereo_params.pickle"))
for pair in camera_pair:
map1_x, map1_y = cv2.initUndistortRectifyMap(
stereo_params[pair[0] + "-" + pair[1]]["cameraMatrix1"],
stereo_params[pair[0] + "-" + pair[1]]["distCoeffs1"],
stereo_params[pair[0] + "-" + pair[1]]["R1"],
stereo_params[pair[0] + "-" + pair[1]]["P1"],
(stereo_params[pair[0] + "-" + pair[1]]["image_shape"][0]),
cv2.CV_16SC2,
)
map2_x, map2_y = cv2.initUndistortRectifyMap(
stereo_params[pair[0] + "-" + pair[1]]["cameraMatrix2"],
stereo_params[pair[0] + "-" + pair[1]]["distCoeffs2"],
stereo_params[pair[0] + "-" + pair[1]]["R2"],
stereo_params[pair[0] + "-" + pair[1]]["P2"],
(stereo_params[pair[0] + "-" + pair[1]]["image_shape"][1]),
cv2.CV_16SC2,
)
cam1_undistort = []
cam2_undistort = []
for fname in images:
if pair[0] in fname:
filename = Path(fname).stem
img1 = cv2.imread(fname)
gray1 = cv2.cvtColor(img1, cv2.COLOR_BGR2GRAY)
h, w = img1.shape[:2]
_, corners1 = cv2.findChessboardCorners(
gray1, (cbcol, cbrow), None)
corners_origin1 = cv2.cornerSubPix(gray1, corners1, (11, 11),
(-1, -1), criteria)
# Remapping dataFrame_camera1_undistort
im_remapped1 = cv2.remap(img1, map1_x, map1_y,
cv2.INTER_LANCZOS4)
imgpoints_proj_undistort = cv2.undistortPoints(
src=corners_origin1,
cameraMatrix=stereo_params[pair[0] + "-" +
pair[1]]["cameraMatrix1"],
distCoeffs=stereo_params[pair[0] + "-" +
pair[1]]["distCoeffs1"],
P=stereo_params[pair[0] + "-" + pair[1]]["P1"],
R=stereo_params[pair[0] + "-" + pair[1]]["R1"],
)
cam1_undistort.append(imgpoints_proj_undistort)
cv2.imwrite(
os.path.join(str(path_undistort),
filename + "_undistort.jpg"),
im_remapped1,
)
imgpoints_proj_undistort = []
elif pair[1] in fname:
filename = Path(fname).stem
img2 = cv2.imread(fname)
gray2 = cv2.cvtColor(img2, cv2.COLOR_BGR2GRAY)
h, w = img2.shape[:2]
_, corners2 = cv2.findChessboardCorners(
gray2, (cbcol, cbrow), None)
corners_origin2 = cv2.cornerSubPix(gray2, corners2, (11, 11),
(-1, -1), criteria)
# Remapping
im_remapped2 = cv2.remap(img2, map2_x, map2_y,
cv2.INTER_LANCZOS4)
imgpoints_proj_undistort2 = cv2.undistortPoints(
src=corners_origin2,
cameraMatrix=stereo_params[pair[0] + "-" +
pair[1]]["cameraMatrix2"],
distCoeffs=stereo_params[pair[0] + "-" +
pair[1]]["distCoeffs2"],
P=stereo_params[pair[0] + "-" + pair[1]]["P2"],
R=stereo_params[pair[0] + "-" + pair[1]]["R2"],
)
cam2_undistort.append(imgpoints_proj_undistort2)
cv2.imwrite(
os.path.join(str(path_undistort),
filename + "_undistort.jpg"),
im_remapped2,
)
imgpoints_proj_undistort2 = []
cam1_undistort = np.array(cam1_undistort)
cam2_undistort = np.array(cam2_undistort)
print("All images are undistorted and stored in %s" %
str(path_undistort))
print(
"Use the function ``triangulate`` to undistort the dataframes and compute the triangulation"
)
if plot == True:
f1, (ax1, ax2) = plt.subplots(1, 2, figsize=(20, 10))
f1.suptitle(
str("Original Image: Views from " + pair[0] + " and " +
pair[1]),
fontsize=25,
)
# Display images in RGB
ax1.imshow(cv2.cvtColor(img1, cv2.COLOR_BGR2RGB))
ax2.imshow(cv2.cvtColor(img2, cv2.COLOR_BGR2RGB))
norm = mcolors.Normalize(vmin=0.0, vmax=cam1_undistort.shape[1])
plt.savefig(os.path.join(str(path_undistort),
"Original_Image.png"))
# Plot the undistorted corner points
f2, (ax1, ax2) = plt.subplots(1, 2, figsize=(20, 10))
f2.suptitle("Undistorted corner points on camera-1 and camera-2",
fontsize=25)
ax1.imshow(cv2.cvtColor(im_remapped1, cv2.COLOR_BGR2RGB))
ax2.imshow(cv2.cvtColor(im_remapped2, cv2.COLOR_BGR2RGB))
for i in range(0, cam1_undistort.shape[1]):
ax1.scatter(
[cam1_undistort[-1][i, 0, 0]],
[cam1_undistort[-1][i, 0, 1]],
marker=markerType,
s=markerSize,
color=markerColor,
alpha=alphaValue,
)
ax2.scatter(
[cam2_undistort[-1][i, 0, 0]],
[cam2_undistort[-1][i, 0, 1]],
marker=markerType,
s=markerSize,
color=markerColor,
alpha=alphaValue,
)
plt.savefig(
os.path.join(str(path_undistort), "undistorted_points.png"))
# Triangulate
triangulate = auxiliaryfunctions_3d.compute_triangulation_calibration_images(
stereo_params[pair[0] + "-" + pair[1]],
cam1_undistort,
cam2_undistort,
path_undistort,
cfg_3d,
plot=True,
)
auxiliaryfunctions.write_pickle("triangulate.pickle", triangulate)
def retrieveData_and_computeEpLines(self, img, imNum):
# load labeledPoints and fundamental Matrix
if self.config3d is not None:
cfg_3d = auxiliaryfunctions.read_config(self.config3d)
cams = cfg_3d["camera_names"]
path_camera_matrix = auxiliaryfunctions_3d.Foldernames3Dproject(
cfg_3d)[2]
path_stereo_file = os.path.join(path_camera_matrix,
"stereo_params.pickle")
stereo_file = auxiliaryfunctions.read_pickle(path_stereo_file)
for cam in cams:
if cam in img:
labelCam = cam
if self.sourceCam is None:
sourceCam = [
otherCam for otherCam in cams
if cam not in otherCam
][0] # WHY?
else:
sourceCam = self.sourceCam
sourceCamIdx = np.where(np.array(cams) == sourceCam)[0][0]
labelCamIdx = np.where(np.array(cams) == labelCam)[0][0]
if sourceCamIdx < labelCamIdx:
camera_pair = cams[sourceCamIdx] + "-" + cams[labelCamIdx]
sourceCam_numInPair = 1
else:
camera_pair = cams[labelCamIdx] + "-" + cams[sourceCamIdx]
sourceCam_numInPair = 2
fundMat = stereo_file[camera_pair]["F"]
sourceCam_path = os.path.split(img.replace(labelCam, sourceCam))[0]
cfg = auxiliaryfunctions.read_config(self.config)
scorer = cfg["scorer"]
try:
dataFrame = pd.read_hdf(
os.path.join(sourceCam_path,
"CollectedData_" + scorer + ".h5"))
dataFrame.sort_index(inplace=True)
except IOError:
print(
"source camera images have not yet been labeled, or you have opened this folder in the wrong mode!"
)
return None, None, None
# Find offset terms for drawing epipolar Lines
# Get crop params for camera being labeled
foundEvent = 0
eventSearch = re.compile(os.path.split(os.path.split(img)[0])[1])
cropPattern = re.compile("[0-9]{1,4}")
with open(self.config, "rt") as config:
for line in config:
if foundEvent == 1:
crop_labelCam = np.int32(re.findall(cropPattern, line))
break
if eventSearch.search(line) != None:
foundEvent = 1
# Get crop params for other camera
foundEvent = 0
eventSearch = re.compile(os.path.split(sourceCam_path)[1])
cropPattern = re.compile("[0-9]{1,4}")
with open(self.config, "rt") as config:
for line in config:
if foundEvent == 1:
crop_sourceCam = np.int32(re.findall(
cropPattern, line))
break
if eventSearch.search(line) != None:
foundEvent = 1
labelCam_offsets = [crop_labelCam[0], crop_labelCam[2]]
sourceCam_offsets = [crop_sourceCam[0], crop_sourceCam[2]]
sourceCam_pts = np.asarray(dataFrame, dtype=np.int32)
sourceCam_pts = sourceCam_pts.reshape(
(sourceCam_pts.shape[0], int(sourceCam_pts.shape[1] / 2), 2))
sourceCam_pts = np.moveaxis(sourceCam_pts, [0, 1, 2], [1, 0, 2])
sourceCam_pts[...,
0] = sourceCam_pts[..., 0] + sourceCam_offsets[0]
sourceCam_pts[...,
1] = sourceCam_pts[..., 1] + sourceCam_offsets[1]
sourcePts = sourceCam_pts[:, imNum, :]
epLines_source2label = cv2.computeCorrespondEpilines(
sourcePts, int(sourceCam_numInPair), fundMat)
epLines_source2label.reshape(-1, 3)
return epLines_source2label, sourcePts, labelCam_offsets
else:
return None, None, None
def triangulate(
config,
video_path,
videotype="",
filterpredictions=True,
filtertype="median",
gputouse=None,
destfolder=None,
save_as_csv=False,
track_method="",
):
"""
This function triangulates the detected DLC-keypoints from the two camera views
using the camera matrices (derived from calibration) to calculate 3D predictions.
Parameters
----------
config : string
Full path of the config.yaml file as a string.
video_path : string/list of list
Full path of the directory where videos are saved. If the user wants to analyze
only a pair of videos, the user needs to pass them as a list of list of videos,
i.e. [['video1-camera-1.avi','video1-camera-2.avi']]
videotype: string, optional
Checks for the extension of the video in case the input to the video is a directory.\n Only videos with this extension are analyzed.
If left unspecified, videos with common extensions ('avi', 'mp4', 'mov', 'mpeg', 'mkv') are kept.
filterpredictions: Bool, optional
Filter the predictions with filter specified by "filtertype". If specified it
should be either ``True`` or ``False``.
filtertype: string
Select which filter, 'arima' or 'median' filter (currently supported).
gputouse: int, optional. Natural number indicating the number of your GPU (see number in nvidia-smi).
If you do not have a GPU put None.
See: https://nvidia.custhelp.com/app/answers/detail/a_id/3751/~/useful-nvidia-smi-queries
destfolder: string, optional
Specifies the destination folder for analysis data (default is the path of the video)
save_as_csv: bool, optional
Saves the predictions in a .csv file. The default is ``False``
Example
-------
Linux/MacOS
To analyze all the videos in the directory:
>>> deeplabcut.triangulate(config,'/data/project1/videos/')
To analyze only a few pairs of videos:
>>> deeplabcut.triangulate(config,[['/data/project1/videos/video1-camera-1.avi','/data/project1/videos/video1-camera-2.avi'],['/data/project1/videos/video2-camera-1.avi','/data/project1/videos/video2-camera-2.avi']])
Windows
To analyze all the videos in the directory:
>>> deeplabcut.triangulate(config,'C:\\yourusername\\rig-95\\Videos')
To analyze only a few pair of videos:
>>> deeplabcut.triangulate(config,[['C:\\yourusername\\rig-95\\Videos\\video1-camera-1.avi','C:\\yourusername\\rig-95\\Videos\\video1-camera-2.avi'],['C:\\yourusername\\rig-95\\Videos\\video2-camera-1.avi','C:\\yourusername\\rig-95\\Videos\\video2-camera-2.avi']])
"""
from deeplabcut.pose_estimation_tensorflow import predict_videos
from deeplabcut.post_processing import filtering
cfg_3d = auxiliaryfunctions.read_config(config)
cam_names = cfg_3d["camera_names"]
pcutoff = cfg_3d["pcutoff"]
scorer_3d = cfg_3d["scorername_3d"]
snapshots = {}
for cam in cam_names:
snapshots[cam] = cfg_3d[str("config_file_" + cam)]
# Check if the config file exists
if not os.path.exists(snapshots[cam]):
raise Exception(
str(
"It seems the file specified in the variable config_file_"
+ str(cam)
)
+ " does not exist. Please edit the config file with correct file path and retry."
)
# flag to check if the video_path variable is a string or a list of list
flag = False # assumes that video path is a list
if isinstance(video_path, str) == True:
flag = True
video_list = auxiliaryfunctions_3d.get_camerawise_videos(
video_path, cam_names, videotype=videotype
)
else:
video_list = video_path
if video_list == []:
print("No videos found in the specified video path.", video_path)
print(
"Please make sure that the video names are specified with correct camera names as entered in the config file or"
)
print(
"perhaps the videotype is distinct from the videos in the path, I was looking for:",
videotype,
)
print("List of pairs:", video_list)
scorer_name = {}
run_triangulate = False
for i in range(len(video_list)):
dataname = []
for j in range(len(video_list[i])): # looping over cameras
if cam_names[j] in video_list[i][j]:
print(
"Analyzing video %s using %s"
% (video_list[i][j], str("config_file_" + cam_names[j]))
)
config_2d = snapshots[cam_names[j]]
cfg = auxiliaryfunctions.read_config(config_2d)
# Get track_method and do related checks
track_method = auxfun_multianimal.get_track_method(
cfg, track_method=track_method
)
if len(cfg.get("multianimalbodyparts", [])) == 1 and track_method != "box":
warnings.warn(
"Switching to `box` tracker for single point tracking..."
)
track_method = "box"
# Get track method suffix
tr_method_suffix = TRACK_METHODS.get(track_method, "")
shuffle = cfg_3d[str("shuffle_" + cam_names[j])]
trainingsetindex = cfg_3d[str("trainingsetindex_" + cam_names[j])]
trainFraction = cfg["TrainingFraction"][trainingsetindex]
if flag == True:
video = os.path.join(video_path, video_list[i][j])
else:
video_path = str(Path(video_list[i][j]).parents[0])
video = os.path.join(video_path, video_list[i][j])
if destfolder is None:
destfolder = str(Path(video).parents[0])
vname = Path(video).stem
prefix = str(vname).split(cam_names[j])[0]
suffix = str(vname).split(cam_names[j])[-1]
if prefix == "":
pass
elif prefix[-1] == "_" or prefix[-1] == "-":
prefix = prefix[:-1]
if suffix == "":
pass
elif suffix[0] == "_" or suffix[0] == "-":
suffix = suffix[1:]
if prefix == "":
output_file = os.path.join(destfolder, suffix)
else:
if suffix == "":
output_file = os.path.join(destfolder, prefix)
else:
output_file = os.path.join(destfolder, prefix + "_" + suffix)
output_filename = os.path.join(
output_file + "_" + scorer_3d
) # Check if the videos are already analyzed for 3d
if os.path.isfile(output_filename + ".h5"):
if save_as_csv is True and not os.path.exists(
output_filename + ".csv"
):
# In case user adds save_as_csv is True after triangulating
pd.read_hdf(output_filename + ".h5").to_csv(
str(output_filename + ".csv")
)
print(
"Already analyzed...Checking the meta data for any change in the camera matrices and/or scorer names",
vname,
)
pickle_file = str(output_filename + "_meta.pickle")
metadata_ = auxiliaryfunctions_3d.LoadMetadata3d(pickle_file)
(
img_path,
path_corners,
path_camera_matrix,
path_undistort,
_,
) = auxiliaryfunctions_3d.Foldernames3Dproject(cfg_3d)
path_stereo_file = os.path.join(
path_camera_matrix, "stereo_params.pickle"
)
stereo_file = auxiliaryfunctions.read_pickle(path_stereo_file)
cam_pair = str(cam_names[0] + "-" + cam_names[1])
is_video_analyzed = False # variable to keep track if the video was already analyzed
# Check for the camera matrix
for k in metadata_["stereo_matrix"].keys():
if np.all(
metadata_["stereo_matrix"][k] == stereo_file[cam_pair][k]
):
pass
else:
run_triangulate = True
# Check for scorer names in the pickle file of 3d output
DLCscorer, DLCscorerlegacy = auxiliaryfunctions.GetScorerName(
cfg, shuffle, trainFraction, trainingsiterations="unknown"
)
if (
metadata_["scorer_name"][cam_names[j]] == DLCscorer
): # TODO: CHECK FOR BOTH?
is_video_analyzed = True
elif metadata_["scorer_name"][cam_names[j]] == DLCscorerlegacy:
is_video_analyzed = True
else:
is_video_analyzed = False
run_triangulate = True
if is_video_analyzed:
print("This file is already analyzed!")
dataname.append(
os.path.join(
destfolder, vname + DLCscorer + tr_method_suffix + ".h5"
)
)
scorer_name[cam_names[j]] = DLCscorer
else:
# Analyze video if score name is different
DLCscorer = predict_videos.analyze_videos(
config_2d,
[video],
videotype=videotype,
shuffle=shuffle,
trainingsetindex=trainingsetindex,
gputouse=gputouse,
destfolder=destfolder,
)
scorer_name[cam_names[j]] = DLCscorer
is_video_analyzed = False
run_triangulate = True
suffix = tr_method_suffix
if filterpredictions:
filtering.filterpredictions(
config_2d,
[video],
videotype=videotype,
shuffle=shuffle,
trainingsetindex=trainingsetindex,
filtertype=filtertype,
destfolder=destfolder,
)
suffix += "_filtered"
dataname.append(
os.path.join(
destfolder, vname + DLCscorer + suffix + ".h5"
)
)
else: # need to do the whole jam.
DLCscorer = predict_videos.analyze_videos(
config_2d,
[video],
videotype=videotype,
shuffle=shuffle,
trainingsetindex=trainingsetindex,
gputouse=gputouse,
destfolder=destfolder,
)
scorer_name[cam_names[j]] = DLCscorer
run_triangulate = True
print(destfolder, vname, DLCscorer)
if filterpredictions:
filtering.filterpredictions(
config_2d,
[video],
videotype=videotype,
shuffle=shuffle,
trainingsetindex=trainingsetindex,
filtertype=filtertype,
destfolder=destfolder,
)
dataname.append(
os.path.join(
destfolder, vname + DLCscorer + tr_method_suffix + "_filtered.h5"
)
)
if run_triangulate:
# if len(dataname)>0:
# undistort points for this pair
print("Undistorting...")
(
dataFrame_camera1_undistort,
dataFrame_camera2_undistort,
stereomatrix,
path_stereo_file,
) = undistort_points(
config, dataname, str(cam_names[0] + "-" + cam_names[1])
)
if len(dataFrame_camera1_undistort) != len(dataFrame_camera2_undistort):
import warnings
warnings.warn(
"The number of frames do not match in the two videos. Please make sure that your videos have same number of frames and then retry! Excluding the extra frames from the longer video."
)
if len(dataFrame_camera1_undistort) > len(dataFrame_camera2_undistort):
dataFrame_camera1_undistort = dataFrame_camera1_undistort[
: len(dataFrame_camera2_undistort)
]
if len(dataFrame_camera2_undistort) > len(dataFrame_camera1_undistort):
dataFrame_camera2_undistort = dataFrame_camera2_undistort[
: len(dataFrame_camera1_undistort)
]
# raise Exception("The number of frames do not match in the two videos. Please make sure that your videos have same number of frames and then retry!")
scorer_cam1 = dataFrame_camera1_undistort.columns.get_level_values(0)[0]
scorer_cam2 = dataFrame_camera2_undistort.columns.get_level_values(0)[0]
scorer_3d = scorer_cam1
bodyparts = dataFrame_camera1_undistort.columns.get_level_values(
"bodyparts"
).unique()
P1 = stereomatrix["P1"]
P2 = stereomatrix["P2"]
F = stereomatrix["F"]
print("Computing the triangulation...")
num_frames = dataFrame_camera1_undistort.shape[0]
### Assign nan to [X,Y] of low likelihood predictions ###
# Convert the data to a np array to easily mask out the low likelihood predictions
data_cam1_tmp = dataFrame_camera1_undistort.to_numpy().reshape(
(num_frames, -1, 3)
)
data_cam2_tmp = dataFrame_camera2_undistort.to_numpy().reshape(
(num_frames, -1, 3)
)
# Assign [X,Y] = nan to low likelihood predictions
data_cam1_tmp[data_cam1_tmp[..., 2] < pcutoff, :2] = np.nan
data_cam2_tmp[data_cam2_tmp[..., 2] < pcutoff, :2] = np.nan
# Reshape data back to original shape
data_cam1_tmp = data_cam1_tmp.reshape(num_frames, -1)
data_cam2_tmp = data_cam2_tmp.reshape(num_frames, -1)
# put data back to the dataframes
dataFrame_camera1_undistort[:] = data_cam1_tmp
dataFrame_camera2_undistort[:] = data_cam2_tmp
if cfg.get("multianimalproject"):
# Check individuals are the same in both views
individuals_view1 = (
dataFrame_camera1_undistort.columns.get_level_values("individuals")
.unique()
.to_list()
)
individuals_view2 = (
dataFrame_camera2_undistort.columns.get_level_values("individuals")
.unique()
.to_list()
)
if individuals_view1 != individuals_view2:
raise ValueError(
"The individuals do not match between the two DataFrames"
)
# Cross-view match individuals
_, voting = auxiliaryfunctions_3d.cross_view_match_dataframes(
dataFrame_camera1_undistort, dataFrame_camera2_undistort, F
)
else:
# Create a dummy variables for single-animal
individuals_view1 = ["indie"]
voting = {0: 0}
# Cleaner variable (since inds view1 == inds view2)
individuals = individuals_view1
# Reshape: (num_framex, num_individuals, num_bodyparts , 2)
all_points_cam1 = dataFrame_camera1_undistort.to_numpy().reshape(
(num_frames, len(individuals), -1, 3)
)[..., :2]
all_points_cam2 = dataFrame_camera2_undistort.to_numpy().reshape(
(num_frames, len(individuals), -1, 3)
)[..., :2]
# Triangulate data
triangulate = []
for i, _ in enumerate(individuals):
# i is individual in view 1
# voting[i] is the matched individual in view 2
pts_indv_cam1 = all_points_cam1[:, i].reshape((-1, 2)).T
pts_indv_cam2 = all_points_cam2[:, voting[i]].reshape((-1, 2)).T
indv_points_3d = auxiliaryfunctions_3d.triangulatePoints(
P1, P2, pts_indv_cam1, pts_indv_cam2
)
indv_points_3d = indv_points_3d[:3].T.reshape((num_frames, -1, 3))
triangulate.append(indv_points_3d)
triangulate = np.asanyarray(triangulate)
metadata = {}
metadata["stereo_matrix"] = stereomatrix
metadata["stereo_matrix_file"] = path_stereo_file
metadata["scorer_name"] = {
cam_names[0]: scorer_name[cam_names[0]],
cam_names[1]: scorer_name[cam_names[1]],
}
# Create 3D DataFrame column and row indices
axis_labels = ("x", "y", "z")
if cfg.get("multianimalproject"):
columns = pd.MultiIndex.from_product(
[[scorer_3d], individuals, bodyparts, axis_labels],
names=["scorer", "individuals", "bodyparts", "coords"],
)
else:
columns = pd.MultiIndex.from_product(
[[scorer_3d], bodyparts, axis_labels],
names=["scorer", "bodyparts", "coords"],
)
inds = range(num_frames)
# Swap num_animals with num_frames axes to ensure well-behaving reshape
triangulate = triangulate.swapaxes(0, 1).reshape((num_frames, -1))
# Fill up 3D dataframe
df_3d = pd.DataFrame(triangulate, columns=columns, index=inds)
df_3d.to_hdf(
str(output_filename + ".h5"),
"df_with_missing",
format="table",
mode="w",
)
# Reorder 2D dataframe in view 2 to match order of view 1
if cfg.get("multianimalproject"):
df_2d_view2 = pd.read_hdf(dataname[1])
individuals_order = [individuals[i] for i in list(voting.values())]
df_2d_view2 = auxfun_multianimal.reorder_individuals_in_df(df_2d_view2, individuals_order)
df_2d_view2.to_hdf(dataname[1], "tracks", format="table", mode="w",)
auxiliaryfunctions_3d.SaveMetadata3d(
str(output_filename + "_meta.pickle"), metadata
)
if save_as_csv:
df_3d.to_csv(str(output_filename + ".csv"))
print("Triangulated data for video", video_list[i])
print("Results are saved under: ", destfolder)
# have to make the dest folder none so that it can be updated for a new pair of videos
if destfolder == str(Path(video).parents[0]):
destfolder = None
if len(video_list) > 0:
print("All videos were analyzed...")
print("Now you can create 3D video(s) using deeplabcut.create_labeled_video_3d")
def undistort_points(config, dataframe, camera_pair, destfolder):
cfg_3d = auxiliaryfunctions.read_config(config)
img_path, path_corners, path_camera_matrix, path_undistort = auxiliaryfunctions_3d.Foldernames3Dproject(
cfg_3d)
'''
path_undistort = destfolder
filename_cam1 = Path(dataframe[0]).stem
filename_cam2 = Path(dataframe[1]).stem
#currently no interm. saving of this due to high speed.
# check if the undistorted files are already present
if os.path.exists(os.path.join(path_undistort,filename_cam1 + '_undistort.h5')) and os.path.exists(os.path.join(path_undistort,filename_cam2 + '_undistort.h5')):
print("The undistorted files are already present at %s" % os.path.join(path_undistort,filename_cam1))
dataFrame_cam1_undistort = pd.read_hdf(os.path.join(path_undistort,filename_cam1 + '_undistort.h5'))
dataFrame_cam2_undistort = pd.read_hdf(os.path.join(path_undistort,filename_cam2 + '_undistort.h5'))
else:
'''
if True:
# Create an empty dataFrame to store the undistorted 2d coordinates and likelihood
dataframe_cam1 = pd.read_hdf(dataframe[0])
dataframe_cam2 = pd.read_hdf(dataframe[1])
scorer_cam1 = dataframe_cam1.columns.get_level_values(0)[0]
scorer_cam2 = dataframe_cam1.columns.get_level_values(0)[0]
stereo_file = auxiliaryfunctions.read_pickle(
os.path.join(path_camera_matrix, 'stereo_params.pickle'))
path_stereo_file = os.path.join(path_camera_matrix,
'stereo_params.pickle')
stereo_file = auxiliaryfunctions.read_pickle(path_stereo_file)
mtx_l = stereo_file[camera_pair]['cameraMatrix1']
dist_l = stereo_file[camera_pair]['distCoeffs1']
mtx_r = stereo_file[camera_pair]['cameraMatrix2']
dist_r = stereo_file[camera_pair]['distCoeffs2']
R1 = stereo_file[camera_pair]['R1']
P1 = stereo_file[camera_pair]['P1']
R2 = stereo_file[camera_pair]['R2']
P2 = stereo_file[camera_pair]['P2']
# Create an empty dataFrame to store the undistorted 2d coordinates and likelihood
dataFrame_cam1_undistort, scorer_cam1, bodyparts = auxiliaryfunctions_3d.create_empty_df(
dataframe_cam1, scorer_cam1, flag='2d')
dataFrame_cam2_undistort, scorer_cam2, bodyparts = auxiliaryfunctions_3d.create_empty_df(
dataframe_cam2, scorer_cam2, flag='2d')
for bpindex, bp in tqdm(enumerate(bodyparts)):
# Undistorting the points from cam1 camera
points_cam1 = np.array([
dataframe_cam1[scorer_cam1][bp]['x'].values[:],
dataframe_cam1[scorer_cam1][bp]['y'].values[:]
])
points_cam1 = points_cam1.T
points_cam1 = np.expand_dims(points_cam1, axis=1)
points_cam1_remapped = cv2.undistortPoints(src=points_cam1,
cameraMatrix=mtx_l,
distCoeffs=dist_l,
P=P1,
R=R1)
dataFrame_cam1_undistort.iloc[:][scorer_cam1, bp,
'x'] = points_cam1_remapped[:, 0,
0]
dataFrame_cam1_undistort.iloc[:][scorer_cam1, bp,
'y'] = points_cam1_remapped[:, 0,
1]
dataFrame_cam1_undistort.iloc[:][
scorer_cam1, bp, 'likelihood'] = dataframe_cam1[scorer_cam1][
bp]['likelihood'].values[:]
# Undistorting the points from cam2 camera
points_cam2 = np.array([
dataframe_cam2[scorer_cam2][bp]['x'].values[:],
dataframe_cam2[scorer_cam2][bp]['y'].values[:]
])
points_cam2 = points_cam2.T
points_cam2 = np.expand_dims(points_cam2, axis=1)
points_cam2_remapped = cv2.undistortPoints(src=points_cam2,
cameraMatrix=mtx_r,
distCoeffs=dist_r,
P=P2,
R=R2)
dataFrame_cam2_undistort.iloc[:][scorer_cam2, bp,
'x'] = points_cam2_remapped[:, 0,
0]
dataFrame_cam2_undistort.iloc[:][scorer_cam2, bp,
'y'] = points_cam2_remapped[:, 0,
1]
dataFrame_cam2_undistort.iloc[:][
scorer_cam2, bp, 'likelihood'] = dataframe_cam2[scorer_cam2][
bp]['likelihood'].values[:]
# Save the undistorted files
dataFrame_cam1_undistort.sort_index(inplace=True)
dataFrame_cam2_undistort.sort_index(inplace=True)
return (dataFrame_cam1_undistort, dataFrame_cam2_undistort,
stereo_file[camera_pair], path_stereo_file)
def triangulate(dlc_3d_config, video_cfg, origins, table_corners, tocchini, video_path, filterpredictions=True, destfolder=None,
save_as_csv=False):
cfg_3d = auxiliaryfunctions.read_config(dlc_3d_config)
img_path, path_corners, path_camera_matrix, path_undistort = auxiliaryfunctions_3d.Foldernames3Dproject(cfg_3d)
cam_names = cfg_3d['camera_names']
pcutoff = cfg_3d['pcutoff']
scorer_3d = cfg_3d['scorername_3d']
snapshots = {}
for cam in cam_names:
snapshots[cam] = cfg_3d[str('config_file_' + cam)]
# Check if the config file exists
if not os.path.exists(snapshots[cam]):
raise Exception(str("It seems the file specified in the variable config_file_" + str(cam)) +
" does not exist. Please edit the config file with correct file path and retry.")
path_stereo_file = os.path.join(path_camera_matrix, 'stereo_params.pickle')
stereo_file = auxiliaryfunctions.read_pickle(path_stereo_file)
position=np.array([0.0,0.0,0.0])
rotation=np.array([[1.0, 0.0, 0.0],[0.0, 1.0, 0.0], [0.0, 0.0, 1.0]])
rotations=[np.linalg.inv(rotation)]
translations=[np.matmul(-rotation, position)]
projections=[]
for cam_idx in range(1,len(cam_names)):
pair_rotation=stereo_file[cam_names[0] + '-' + cam_names[cam_idx]]['R']
pair_translation = stereo_file[cam_names[0] + '-' + cam_names[cam_idx]]['T']
rotations.append(np.matmul(pair_rotation,rotations[0]))
translations.append(np.matmul(pair_rotation, translations[0])+np.transpose(pair_translation))
for cam_idx in range(len(cam_names)):
path_camera_file = os.path.join(path_camera_matrix, '%s_intrinsic_params.pickle' % cam_names[cam_idx])
intrinsic_file = auxiliaryfunctions.read_pickle(path_camera_file)
projection=np.zeros((3,4))
projection[:,0:3]=rotations[cam_idx]
projection[:,3]=translations[cam_idx]
projection=np.matmul(intrinsic_file[cam_names[cam_idx]]['mtx'], projection)
projections.append(projection)
for view in cam_names:
origins[view]=np.array(origins[view])
origins[view][0]=origins[view][0]+video_cfg['crop_limits'][view][0]
origins[view][1]=origins[view][1]+video_cfg['crop_limits'][view][2]
[origin, pairs_used]=locate(cam_names,{'front':1,'side':1,'top':1}, origins, pcutoff, projections, reconMode='all')
table_coords_3d=[]
for corner_idx in range(len(table_corners['front'])):
corner={}
for view in cam_names:
corner[view]=table_corners[view][corner_idx]
corner[view] =np.array(corner[view])
corner[view][0]=corner[view][0]+video_cfg['crop_limits'][view][0]
corner[view][1]=corner[view][1]+video_cfg['crop_limits'][view][2]
[coord,pairs_used]=locate(cam_names,{'front':1,'side':1,'top':1},corner,pcutoff,projections, reconMode='all')
table_coords_3d.append(coord)
xy_norm=np.array([[0,0,1]])
table_center=np.mean(np.array(table_coords_3d),axis=0)
table_vec1=table_coords_3d[0]-table_center
table_vec2=table_coords_3d[1]-table_center
table_norm = unit_vector(np.cross(np.transpose(table_vec1), np.transpose(table_vec2)))
rot_vec=unit_vector(np.cross(xy_norm, table_norm), axis=1)
rot_angle=-np.arccos(np.abs(np.sum(xy_norm*table_norm))/np.sqrt(np.sum(table_norm**2)))
rot_mat=rotation_matrix(rot_angle, np.transpose(rot_vec))
rot_mat=np.matmul(rotation_matrix(np.pi, [1, 0, 0]), rot_mat)
rot_mat=rot_mat[0:3,0:3]
origin=np.matmul(rot_mat, origin)
for idx, coord in enumerate(table_coords_3d):
coord=np.matmul(rot_mat, coord)-origin
table_coords_3d[idx]=coord
tocchini_coords_3d=[]
for tocchino_idx in range(len(tocchini['front'])):
tocchino={}
for view in cam_names:
tocchino[view]=tocchini[view][tocchino_idx]
tocchino[view]=np.array(tocchino[view])
tocchino[view][0]=tocchino[view][0]+video_cfg['crop_limits'][view][0]
tocchino[view][1]=tocchino[view][1]+video_cfg['crop_limits'][view][2]
[coord,pairs_used]=locate(cam_names,{'front':1,'side':1,'top':1},tocchino,pcutoff,projections, reconMode='all')
tocchino_coord=np.matmul(rot_mat,coord)-origin
tocchini_coords_3d.append(tocchino_coord)
file_name_3d_scorer = []
dataname = []
for cam_name in cam_names:
dlc_3d_config = snapshots[cam_name]
cfg = auxiliaryfunctions.read_config(dlc_3d_config)
shuffle = cfg_3d[str('shuffle_' + cam_name)]
trainingsetindex = cfg_3d[str('trainingsetindex_' + cam_name)]
video=video_path[cam_name]
vname = Path(video).stem
trainFraction = cfg['TrainingFraction'][trainingsetindex]
modelfolder = os.path.join(cfg["project_path"],
str(auxiliaryfunctions.GetModelFolder(trainFraction, shuffle, cfg)))
path_test_config = Path(modelfolder) / 'test' / 'pose_cfg.yaml'
dlc_cfg = load_config(str(path_test_config))
Snapshots = np.array([fn.split('.')[0] for fn in os.listdir(os.path.join(modelfolder, 'train')) if "index" in fn])
snapshotindex = cfg['snapshotindex']
increasing_indices = np.argsort([int(m.split('-')[1]) for m in Snapshots])
Snapshots = Snapshots[increasing_indices]
dlc_cfg['init_weights'] = os.path.join(modelfolder, 'train', Snapshots[snapshotindex])
trainingsiterations = (dlc_cfg['init_weights'].split(os.sep)[-1]).split('-')[-1]
DLCscorer = auxiliaryfunctions.GetScorerName(cfg, shuffle, trainFraction,
trainingsiterations=trainingsiterations)
file_name_3d_scorer.append(DLCscorer)
if filterpredictions:
dataname.append(os.path.join(destfolder, cam_name, vname + DLCscorer + 'filtered.h5'))
else:
dataname.append(os.path.join(destfolder, cam_name, vname + DLCscorer + '.h5'))
output_filename = os.path.join(destfolder, vname + '_' + scorer_3d)
if os.path.isfile(output_filename + '.h5'): # TODO: don't check twice and load the pickle file to check if the same snapshots + camera matrices were used.
print("Already analyzed...", output_filename+'.h5')
else:
if len(dataname) > 0:
df = pd.read_hdf(dataname[0])
df_3d, scorer_3d, bodyparts = auxiliaryfunctions_3d.create_empty_df(df, scorer_3d, flag='3d')
for bpindex, bp in enumerate(bodyparts):
bp_coords=np.zeros((3,len(df_3d)))
for f_idx in range(len(df_3d)):
likelihoods={}
coords={}
for cam_idx,cam_name in enumerate(cam_names):
dataframe_cam = pd.read_hdf(dataname[cam_idx])
scorer_cam = dataframe_cam.columns.get_level_values(0)[0]
likelihoods[cam_name]=dataframe_cam[scorer_cam][bp]['likelihood'].values[f_idx]
coords[cam_name]=np.array([dataframe_cam[scorer_cam][bp]['x'].values[f_idx], dataframe_cam[scorer_cam][bp]['y'].values[f_idx]])
coords[cam_name][0]=coords[cam_name][0]+video_cfg['crop_limits'][cam_name][0]
coords[cam_name][1]=coords[cam_name][1]+video_cfg['crop_limits'][cam_name][2]
[coord, pairs_used] = locate(cam_names, likelihoods, coords, pcutoff, projections, reconMode='bestpossible')
coord=np.matmul(rot_mat, coord)-origin
if pairs_used < 3:
coord[0]=np.nan
coord[1]=np.nan
coord[2]=np.nan
bp_coords[:,f_idx]=np.squeeze(coord)
df_3d.iloc[:][scorer_3d, bp, 'x'] = bp_coords[0,:]
df_3d.iloc[:][scorer_3d, bp, 'y'] = bp_coords[1,:]
df_3d.iloc[:][scorer_3d, bp, 'z'] = bp_coords[2,:]
df_3d.to_hdf(str(output_filename + '.h5'), 'df_with_missing', format='table', mode='w')
if save_as_csv:
df_3d.to_csv(str(output_filename + '.csv'))
print("Triangulated data for video", vname)
print("Results are saved under: ", destfolder)
return str(output_filename + '.h5'), table_coords_3d, tocchini_coords_3d
def align_videos(subject, date):
frame_buffer = 10
base_video_path=os.path.join(cfg['preprocessed_data_dir'],subject, date,'video')
with open(os.path.join(base_video_path,'config.json')) as json_file:
vid_cfg=json.load(json_file)
fnames=sorted(glob.glob(os.path.join(base_video_path, 'front', '%s*.avi' % date)))
rois_checked={
'front':False,
'side':False,
'top':False
}
crop_checked={
'front':False,
'side':False,
'top':False
}
# For each file (filenames are same in each view directory)
for fname in fnames:
fname=os.path.split(fname)[-1]
print('Processing %s' % fname)
blue_onsets={}
yellow_onsets={}
blue_ts={}
yellow_ts={}
video_nframes={}
# Whether or not to use LED for alignment
led_based = True
for view in cfg['camera_views']:
video_path=os.path.join(base_video_path, view)
clip = VideoFileClip(os.path.join(video_path, fname))
n_frames_approx = int(np.ceil(clip.duration * clip.fps) + frame_buffer)
n_frames = n_frames_approx
clip.reader.initialize()
# Initialize LED time series for this view
blue_ts[view]=[]
yellow_ts[view]=[]
for index in range(n_frames_approx):
image = img_as_ubyte(clip.reader.read_frame())
# If not already set, show GUI to select blue LED ROI
if not rois_checked[view]:
blue_led_roi_area=vid_cfg['blue_led_roi_areas'][view]
blue_cropped_img=image[blue_led_roi_area[2]:blue_led_roi_area[3],
blue_led_roi_area[0]:blue_led_roi_area[1],:]
init_roi=None
if view in vid_cfg['blue_led_rois'] and vid_cfg['blue_led_rois'][view] is not None:
init_roi=vid_cfg['blue_led_rois'][view]
init_roi[0] = init_roi[0] - blue_led_roi_area[0]
init_roi[1] = init_roi[1] - blue_led_roi_area[0]
init_roi[2] = init_roi[2] - blue_led_roi_area[2]
init_roi[3] = init_roi[3] - blue_led_roi_area[2]
vid_cfg['blue_led_rois'][view] = select_crop_parameters.show(blue_cropped_img, 'Select blue LED ROI', init_coords=init_roi)
vid_cfg['blue_led_rois'][view][0] = vid_cfg['blue_led_rois'][view][0] + blue_led_roi_area[0]
vid_cfg['blue_led_rois'][view][1] = vid_cfg['blue_led_rois'][view][1] + blue_led_roi_area[0]
vid_cfg['blue_led_rois'][view][2] = vid_cfg['blue_led_rois'][view][2] + blue_led_roi_area[2]
vid_cfg['blue_led_rois'][view][3] = vid_cfg['blue_led_rois'][view][3] + blue_led_roi_area[2]
yellow_led_roi_area = vid_cfg['yellow_led_roi_areas'][view]
yellow_cropped_img = image[yellow_led_roi_area[2]:yellow_led_roi_area[3],
yellow_led_roi_area[0]:yellow_led_roi_area[1], :]
init_roi = None
if view in vid_cfg['yellow_led_rois'] and vid_cfg['yellow_led_rois'][view] is not None:
init_roi = vid_cfg['yellow_led_rois'][view]
init_roi[0] = init_roi[0] - yellow_led_roi_area[0]
init_roi[1] = init_roi[1] - yellow_led_roi_area[0]
init_roi[2] = init_roi[2] - yellow_led_roi_area[2]
init_roi[3] = init_roi[3] - yellow_led_roi_area[2]
vid_cfg['yellow_led_rois'][view] = select_crop_parameters.show(yellow_cropped_img, 'Select yellow LED ROI', init_coords=init_roi)
vid_cfg['yellow_led_rois'][view][0] = vid_cfg['yellow_led_rois'][view][0] + yellow_led_roi_area[0]
vid_cfg['yellow_led_rois'][view][1] = vid_cfg['yellow_led_rois'][view][1] + yellow_led_roi_area[0]
vid_cfg['yellow_led_rois'][view][2] = vid_cfg['yellow_led_rois'][view][2] + yellow_led_roi_area[2]
vid_cfg['yellow_led_rois'][view][3] = vid_cfg['yellow_led_rois'][view][3] + yellow_led_roi_area[2]
rois_checked[view]=True
if index == int(n_frames_approx - frame_buffer * 2):
last_image = image
elif index > int(n_frames_approx - frame_buffer * 2):
if (image == last_image).all():
n_frames = index
break
# Crop image around blue LED, get only blue channel
blue_roi=vid_cfg['blue_led_rois'][view]
blue_led_image = image[blue_roi[2]:blue_roi[3], blue_roi[0]:blue_roi[1], 2]
# Add average of cropped image to blue LED timeseries
blue_ts[view].append(np.mean(blue_led_image))
# Crop image around yellow LED, average red and green channels
yellow_roi=vid_cfg['yellow_led_rois'][view]
yellow_led_image = np.mean(image[yellow_roi[2]:yellow_roi[3], yellow_roi[0]:yellow_roi[1], 0:1],axis=2)
# Add average of cropped image to yellow LED timeseries
yellow_ts[view].append(np.mean(yellow_led_image))
blue_ts[view] = np.array(blue_ts[view])
yellow_ts[view] = np.array(yellow_ts[view])
# Normalize based on first 10 time steps
if len(blue_ts[view])>10:
blue_ts[view]=(blue_ts[view]-np.mean(blue_ts[view][0:10]))/np.mean(blue_ts[view][0:10])
if len(yellow_ts[view])>10:
yellow_ts[view]=(yellow_ts[view]-np.mean(yellow_ts[view][0:10]))/np.mean(yellow_ts[view][0:10])
blue_ts[view]=blue_ts[view]/np.max(blue_ts[view])
yellow_ts[view]=yellow_ts[view]/np.max(yellow_ts[view])
# plt.figure()
# plt.subplot(2,1,1)
# plt.plot(video_blue_brightness)
# plt.subplot(2, 1, 2)
# plt.plot(video_yellow_brightness)
# plt.show()
# Get derivative of blue and yellow ts
#blue_diff=np.diff(blue_ts[view])
#yellow_diff=np.diff(yellow_ts[view])
# Get peak blue and yellow LED change times
#blue_peak=np.max(blue_diff)
blue_peak = np.max(blue_ts[view])
#yellow_peak=np.max(yellow_diff)
yellow_peak = np.max(yellow_ts[view])
# If none above 0.05, don't use LEDs for aligning
#if blue_peak<.05 or yellow_peak<.05:
if len(blue_ts[view])<10 or np.max(blue_ts[view][10:]) < .25 or np.max(yellow_ts[view])<.25:
led_based=False
print('Cant figure out LED onset - not using')
# Otherwise, use the first time point after 25 time points where LED diff exceeds 0.05
else:
#blue_onsets[view] = np.where(blue_diff >= 0.05)[0][0]
#np.where(blue_ts[view] >= 0.25)[0][0]
blue_onsets[view] = 10 + np.where(blue_ts[view][10:] >= 0.25)[0][0]
#yellow_onsets[view] = np.where(yellow_diff >= 0.05)[0][0]
yellow_onsets[view] = np.where(yellow_ts[view] >= 0.25)[0][0]
video_nframes[view]=n_frames
# Use first view where blue LED exceeds threshold as reference to align to
if len(blue_onsets.values())>0:
min_blue_onset=min(blue_onsets.values())
# if fname=='15-05-2019_10-34-15_11.avi':
# plt.figure()
# for view in cfg['camera_views']:
# plt.plot(blue_ts[view], label='%s: blue' % view)
# plt.plot(yellow_ts[view], label='%s: yellow' % view)
# plt.legend()
# plt.show()
# Compute trial duration based on each view
trial_durations={}
for view in cfg['camera_views']:
if view in blue_onsets and view in yellow_onsets:
# Trial duration (in ms)
trial_duration=(yellow_onsets[view]-blue_onsets[view])*1.0/clip.fps*1000.0
# there is an 850ms delay before blue LED comes on
if trial_duration>0:
trial_duration=trial_duration+850.0
trial_durations[view]=trial_duration
print('%s: %.2fms' % (view, trial_duration))
#assert(len(trial_durations)>0 and all(x == trial_durations[0] for x in trial_durations))
start_frames_to_cut={}
n_frames_after_cutting = {}
# Cut frames to align videos and crop
for idx,view in enumerate(cfg['camera_views']):
# using LED to align
if led_based:
start_frames_to_cut[view]=blue_onsets[view]-min_blue_onset
# otherwise - use standard # of frames to crop (order of video triggering is top, side, front)
if not led_based or start_frames_to_cut[view]>5:
start_frames_to_cut[view] = 0
if view=='front':
start_frames_to_cut[view]=2
elif view=='side':
start_frames_to_cut[view]=1
n_frames_after_cutting[view]=video_nframes[view]-start_frames_to_cut[view]
new_nframes=min(n_frames_after_cutting.values())
intrinsic_files = {}
for view in cfg['camera_views']:
dlc3d_cfg = os.path.join('/data/tool_learning/preprocessed_data/dlc_projects',
'visual_grasp_3d-Jimmy-2019-08-19-3d', 'config.yaml')
cfg_3d = auxiliaryfunctions.read_config(dlc3d_cfg)
img_path, path_corners, path_camera_matrix, path_undistort = auxiliaryfunctions_3d.Foldernames3Dproject(
cfg_3d)
path_intrinsic_file = os.path.join(path_camera_matrix, '%s_intrinsic_params.pickle' % view)
intrinsic_file = auxiliaryfunctions.read_pickle(path_intrinsic_file)
intrinsic_files[view] = intrinsic_file[view]
for idx, view in enumerate(cfg['camera_views']):
camera_matrix = intrinsic_files[view]['mtx']
distortion_coefficients = intrinsic_files[view]['dist']
end_frames_to_cut=n_frames_after_cutting[view]-new_nframes
print('cutting %d frames from beginning and %d frames from end of %s' % (start_frames_to_cut[view], end_frames_to_cut, view))
# Cut frames from blue and yellow LED time series and onsets
if end_frames_to_cut>0:
blue_ts[view]=blue_ts[view][start_frames_to_cut[view]:-end_frames_to_cut]
yellow_ts[view] = yellow_ts[view][start_frames_to_cut[view]:-end_frames_to_cut]
else:
blue_ts[view] = blue_ts[view][start_frames_to_cut[view]:]
yellow_ts[view] = yellow_ts[view][start_frames_to_cut[view]:]
if view in blue_onsets:
blue_onsets[view]=blue_onsets[view]-start_frames_to_cut[view]
if view in yellow_onsets:
yellow_onsets[view]=yellow_onsets[view]-start_frames_to_cut[view]
# Load video and cut frames from beginning
video_path = os.path.join(base_video_path, view)
clip = VideoFileClip(os.path.join(video_path, fname))
# Crop limits based on view
frames=[]
n_frames_approx = int(np.ceil(clip.duration * clip.fps)+frame_buffer)
for index in range(n_frames_approx):
image = img_as_ubyte(clip.reader.read_frame())
image = cv2.undistort(image, camera_matrix, distortion_coefficients)
if index>=start_frames_to_cut[view]:
if not crop_checked[view]:
init_crop_lims = None
if view in vid_cfg['crop_limits'] and vid_cfg['crop_limits'][view] is not None:
init_crop_lims = vid_cfg['crop_limits'][view]
vid_cfg['crop_limits'][view] = select_crop_parameters.show(image, 'Select crop limits',
init_coords=init_crop_lims)
crop_checked[view]=True
# Crop image and save to video
crop_lims=vid_cfg['crop_limits'][view]
image=image[crop_lims[2]:crop_lims[3], crop_lims[0]:crop_lims[1], :]
frames.append(image)
if len(frames)==new_nframes:
break
clip.close()
# Check that have the right number of frames
assert(len(frames)==new_nframes)
# Create new video clip (cropped and aligned)
video_path = os.path.join(base_video_path, view)
new_clip = VideoProcessorCV(sname=os.path.join(video_path, fname), fps=clip.fps, codec='mp4v',
sw=crop_lims[1] - crop_lims[0], sh=crop_lims[3] - crop_lims[2])
for frame in frames:
new_clip.save_frame(np.uint8(frame))
new_clip.close()
# Make everything hashable
for view in cfg['camera_views']:
blue_ts[view]=blue_ts[view].tolist()
yellow_ts[view] = yellow_ts[view].tolist()
if view in blue_onsets:
blue_onsets[view]=int(blue_onsets[view])
if view in yellow_onsets:
yellow_onsets[view]=int(yellow_onsets[view])
if view in trial_durations:
trial_durations[view]=float(trial_durations[view])
# Save video info to JSON
data = {
'blue_roi': vid_cfg['blue_led_rois'],
'yellow_roi': vid_cfg['yellow_led_rois'],
'blue_ts': blue_ts,
'yellow_ts': yellow_ts,
'blue_onset': blue_onsets,
'yellow_onset': yellow_onsets,
'trial_duration': trial_durations,
'fname': fname
}
[base, ext] = os.path.splitext(fname)
with open(os.path.join(base_video_path, '%s.json' % base), 'w') as outfile:
json.dump(data, outfile)
print('')
with open(os.path.join(base_video_path,'config.json'),'w') as outfile:
json.dump(vid_cfg, outfile)
