def get_DLC_prediction(file_path = None,
dlc_config_file = r"C:\Users\asree\Downloads\Tool_tracking_main\Tool_Tracking-Ambareesh-2019-07-31\config.yaml",
threshold=0.0,
display = False, marker_size = 10, save_path = None, save_format='png'):
"""
filepath: Complete file.path of the image to be tested
threshold: The probability value for the prediction points. If a point is below threshold it won't be marked
display: If True display the test and final images
#TO DO:
Remove boundary
Preserve size as is
"""
import pathlib
import matplotlib.pyplot as plt
import pandas as pd
import deeplabcut
path_config_file = dlc_config_file
directory_path = str(pathlib.Path(file_path).parent)
file_suffix = pathlib.Path(file_path).suffix
deeplabcut.analyze_time_lapse_frames(path_config_file, directory=directory_path, frametype=file_suffix, save_as_csv=True) #looks promising
im = plt.imread(file_path)[:,:,:3] #.PNG can have four channels!
for file_name in pathlib.Path(file_path).parent.iterdir():
if '.csv' in str(file_name):
pred_csv = pd.read_csv(file_name, skiprows=[0,1], usecols=lambda x: x not in 'coords')
X, Y = [], []
for x, y, prob in zip(*[iter(pred_csv.columns)]*3):
if pred_csv.loc[0,prob] > threshold:
X.append(pred_csv.loc[0,x])
Y.append(pred_csv.loc[0,y])
plt.figure(figsize=(im.shape[0]/96, im.shape[1]/96)) # CHANGE TO INCORPORATE CORRECT SIZE
implot = plt.imshow(im.copy())
plt.scatter(x= X, y= Y, c= 'b', s=marker_size)
if not save_path:
plt.savefig('dlc_prediction.'+save_format, bbox_inches='tight', dpi = 'figure')
else:
plt.savefig(pathlib.Path(save_path).joinpath('dlc_prediction.'+save_format), bbox_inches='tight', dpi = 'figure')
if not display:
plt.close()
print(f"DLC prediction done, file saved as dlc_prediction.{save_format}")
break
Shuffles=[2],
net_type=net_type,
augmenter_type=augmenter_type3)
posefile = os.path.join(
cfg['project_path'], 'dlc-models/iteration-' + str(cfg['iteration']) +
'/' + cfg['Task'] + cfg['date'] + '-trainset' +
str(int(cfg['TrainingFraction'][0] * 100)) + 'shuffle' + str(2),
'train/pose_cfg.yaml')
DLC_config = deeplabcut.auxiliaryfunctions.read_plainconfig(posefile)
DLC_config['save_iters'] = 10
DLC_config['display_iters'] = 2
DLC_config['multi_step'] = [[0.001, 10]]
print("CHANGING training parameters to end quickly!")
deeplabcut.auxiliaryfunctions.write_plainconfig(posefile, DLC_config)
print("TRAINING shuffle 2, with smaller allocated memory")
deeplabcut.train_network(path_config_file, shuffle=2, allow_growth=True)
print("ANALYZING some individual frames")
deeplabcut.analyze_time_lapse_frames(
path_config_file,
os.path.join(cfg['project_path'], 'labeled-data/reachingvideo1/'))
print("ALL DONE!!! - default cases are functional.")
print("Re-import DLC with env. variable set to test DLC light mode.")
os.environ['DLClight'] = 'True'
subprocess.call(['python3', "-c", "import deeplabcut"])
def moving_window(x_1, y_1, x_2, y_2, stp, path_img_big, path_image_tree,
config_path):
textFile_filter_results = path_image_tree + 'results_image_filter.txt'
## filePath = String(date) + '_' + String(time) + '_datalog.txt'
## File filePath = SD.open (filePath , FILE_WRITE)
## filePath.println('date, hour, min, sec, T1, R1, T2, R2, R, G, B, L')
##
textFile_filter_results = open(textFile_filter_results, 'w')
textFile_filter_results.write(
'y_pixel,n_trees,n,n_caught_by_dlc_prob,n_caught_by_slope,n_caught_by_array_error,n_caught_by_tree_pixels\n'
)
x_rows = range(x_1, x_2, stp)
y_rows = range(y_1, y_2, stp)
for ypixel in y_rows:
if 'img' in locals():
img.close()
print('closed img')
Image.MAX_IMAGE_PIXELS = 265568257
img = PIL.Image.open(path_img_big)
img.load()
path_im_crop_yrow = path_image_tree + str(ypixel) + '/'
if os.path.exists(path_im_crop_yrow):
print(path_im_crop_yrow + ' exists')
else:
print(path_im_crop_yrow + ' made')
os.mkdir(path_im_crop_yrow)
textFile = path_im_crop_yrow + 'predicted_tree_coordinates.txt'
textFile = open(textFile, 'w')
textFile.write(
'raster,x_1,y_1,x_2,y_2,slope,intercept,tree_px,delta_tree,image_px\n'
)
cropped_images = []
for xpixel in x_rows:
left = xpixel
upper = ypixel - 50
right = xpixel + 50
lower = ypixel
path_im_crop = path_im_crop_yrow + str(left) + '_' + str(
upper) + '_' + str(right) + '_' + str(lower) + '.png'
cropped_images.append(path_im_crop)
if os.path.exists(path_im_crop):
print(path_im_crop + ' exists')
else:
im_crop = img.crop((left, upper, right, lower))
im_crop.save(path_im_crop)
im_crop.close()
img.close()
deeplabcut.analyze_time_lapse_frames(config_path,
path_im_crop_yrow,
frametype='.png',
shuffle=2,
save_as_csv=True)
path_dlc_csv = glob.glob(path_im_crop_yrow + '*.csv')
path_dlc_h5 = glob.glob(path_im_crop_yrow +
'*.h5') ## costs a lot of storage space
path_dlc_pickle = glob.glob(
path_im_crop_yrow + '*.pickle') ## costs a lot of storage space
## we want to get rid of both the .pickle and .h5 files since we will not be using them
for h5 in path_dlc_h5:
os.remove(h5)
for pickle in path_dlc_pickle:
os.remove(pickle)
dat = pandas.read_csv(
path_dlc_csv[0]) ## opens up the .csv file created by dlc
os.remove(path_dlc_csv[0])
dat = dat.iloc[
2:] # gets rid of the first two rows of dat, which are strings and will cause erros
n_caught_by_dlc_prob = 0
n_caught_by_slope = 0
n_caught_by_array_error = 0
n_caught_by_tree_pixels = 0
n_trees = 0
n = 0
for row in dat.iterrows():
n = n + 1
crap, data = row
image_dat = data[0]
x_b = float(data[1])
y_b = float(data[2])
p1l = float(data[3])
x_t = float(data[4])
y_t = float(data[5])
p2l = float(data[6])
if (x_t - x_b) != 0:
slope = (y_t - y_b) / (x_t - x_b)
else:
slope = 0
intercept = y_b - slope * x_b
path_img = path_im_crop_yrow + image_dat
if p1l >= 0.9 and p2l >= 0.9:
if slope > -2 and slope < -0.5:
img = Image.open(path_img).convert('L')
m = np.array(img)
x_min = round(x_b, 0)
x_max = round(x_t, 0)
m_mean = np.mean(m)
pixel_values = []
if x_min < 50 and x_max < 50:
for i in range(int(x_min), int(x_max)):
pixel_i = slope * i + intercept
pixel_i = round(pixel_i, 0)
if pixel_i < 50:
pixel_values.append(m[int(pixel_i), i])
if len(pixel_values) > 0:
tree_px = float(
sum(pixel_values) / len(pixel_values))
delta_tree = m_mean - tree_px
if tree_px is not None and tree_px > 45 and tree_px < 114 and delta_tree > 42 and delta_tree < 137:
n_trees = n_trees + 1
img.close()
os.remove(path_img)
'''
img = Image.open(path_img)
draw = ImageDraw.Draw(img)
shape_line = [(x_b,y_b),(x_t,y_t)]
draw.line(shape_line, fill = 'blue', width = 0)
draw.point((x_b, y_b), fill = 'red')
draw.point((x_t, y_t), fill = 'red')
img.save(path_img)
img.close()
'''
textFile.write(
os.path.basename(path_img).split('.')[0] + ',')
textFile.write(
str(x_b) + ',' + str(y_b) + ',' + str(x_t) +
',' + str(y_t) + ',' + str(slope) + ',')
textFile.write(
str(intercept) + ',' + str(tree_px) + ',')
textFile.write(
str(delta_tree) + ',' + str(m_mean) + '\n')
else:
n_caught_by_tree_pixels = n_caught_by_tree_pixels + 1
img.close()
os.remove(path_img)
'''
path_img = path_im_crop_yrow + image_dat
img = Image.open(path_img)
draw = ImageDraw.Draw(img)
shape_line = [(x_b,y_b),(x_t,y_t)]
draw.line(shape_line, fill = 'green', width = 0)
draw.point((x_b, y_b), fill = 'cyan')
draw.point((x_t, y_t), fill = 'cyan')
img.save(path_img)
img.close()
'''
else:
n_caught_by_array_error = n_caught_by_array_error + 1
img.close()
os.remove(path_img)
else:
n_caught_by_slope = n_caught_by_slope + 1
img.close()
os.remove(path_img)
'''
path_img = path_im_crop_yrow + image_dat
img = Image.open(path_img)
draw = ImageDraw.Draw(img)
shape_line = [(x_b,y_b),(x_t,y_t)]
draw.line(shape_line, fill = 'orange', width = 0)
draw.point((x_b, y_b), fill = 'yellow')
draw.point((x_t, y_t), fill = 'yellow')
img.save(path_img)
img.close()
'''
else:
n_caught_by_dlc_prob = n_caught_by_dlc_prob + 1
os.remove(path_img)
'''
path_img = path_im_crop_yrow + image_dat
img = Image.open(path_img)
draw = ImageDraw.Draw(img)
shape_line = [(x_b,y_b),(x_t,y_t)]
draw.line(shape_line, fill = 'red', width = 0)
draw.point((x_b, y_b), fill = 'cyan')
draw.point((x_t, y_t), fill = 'cyan')
img.save(path_img)
img.close()
'''
textFile.close()
textFile_filter_results.write(str(ypixel) + ',')
textFile_filter_results.write(str(n_trees) + ',')
textFile_filter_results.write(str(n) + ',')
textFile_filter_results.write(str(n_caught_by_dlc_prob) + ',')
textFile_filter_results.write(str(n_caught_by_slope) + ',')
textFile_filter_results.write(str(n_caught_by_array_error) + ',')
textFile_filter_results.write(str(n_caught_by_tree_pixels) + '\n')
textFile_filter_results.close()
)
path_config = '/home/weertman/Documents/Theresa_FHL/Armina_DeepLabCut/Arimina_model_1/'
if os.path.exists(path_config) != True:
os.mkdir(path_config)
path_config = deeplabcut.create_new_project(project,
experimenter,
videos,
copy_videos=True,
working_directory=path_config)
path_config = '/home/weertman/Documents/Theresa_FHL/Armina_DeepLabCut/Arimina_model_1/Armina_model_1-Theresa-2020-07-01/config.yaml'
'''
deeplabcut.extract_frames(path_config,
mode = 'automatic',
algo = 'uniform',
userfeedback = False,
crop = False)
'''
deeplabcut.label_frames(path_config)
deeplabcut.create_training_dataset(path_config)
deeplabcut.train_network(path_config, maxiters=200000, displayiters=10)
deeplabcut.evaluate_network(path_config)
deeplabcut.analyze_time_lapse_frames(
path_config,
'/home/weertman/Documents/Theresa_FHL/Armina_DeepLabCut/left_test_images/',
save_as_csv=True)
# saveDirMovie = '/home/felix/biteData/FFDEET/test/'
img_array = []
for filename in files:
img = cv2.imread(filename)
height, width, layers = img.shape
size = (width, height)
img_array.append(img)
videoBaseName = os.path.basename(os.path.normpath(dataDir))
if '.' in videoBaseName:
videoBaseName = videoBaseName[:-2]
videoName = dataDir + videoBaseName + '.avi'
out = cv2.VideoWriter(videoName, cv2.VideoWriter_fourcc(*'DIVX'), 25, size)
for i in range(len(img_array)):
out.write(img_array[i])
out.release()
videoPath = videoName
imageDir = dataDir
deeplabcut.analyze_time_lapse_frames(config,
imageDir,
save_as_csv=True,
rgb=False)
deeplabcut.create_labeled_video(config, [videoPath], filtered=False)
@author: viviani
"""
import deeplabcut as dlc
from accdatatools.DeepLabCutPipeline.cmdline_utils import config_path as machine_learning_config_path
#Deeplabcut doesn't have inbuilt tools for making seperate testing
#datasets...so we're going to have to hack something together.
#The authors have provided some instruction for how to make this happen
#see here: https://forum.image.sc/t/how-to-move-partial-labeled-frame-to-a-new-project-how-to-quickly-evaluate-a-trained-dataset-on-analyzing-new-frames-without-retraining/29793
#and here: https://forum.image.sc/t/is-there-a-way-to-evaluate-a-deeplabcut-network-on-data-not-present-in-the-test-training-set/32222
video_list = ["C:/Users/viviani/Desktop/2017-02-08_01_CFEB040_eye.mp4"]
testing_config = "C:\\Users\\viviani\\Desktop\\testing_dataset_factory-viviani-2020-07-27\\config.yaml"
if __name__=="__main__":
# dlc.create_new_project("testing_dataset_factory",
# "viviani",
# videos = video_list,
# working_directory = "C:/Users/viviani/Desktop",
# copy_videos=True)
# dlc.extract_frames(testing_config, mode='automatic', algo='kmeans', crop=False)
dlc.label_frames(testing_config)
dlc.analyze_time_lapse_frames(machine_learning_config_path,
"C:/Users/viviani/Desktop/testing_dataset_factory-viviani-2020-07-27/labeled-data/2017-02-08_01_CFEB040_eye",
save_as_csv=True)
def moving_window(x_1, y_1, x_2, y_2, stp, path_img_big, big_img_size,
path_image_tree, config_path):
textFile_filter_results = path_image_tree + 'results_image_filter.txt'
## filePath = String(date) + '_' + String(time) + '_datalog.txt'
## File filePath = SD.open (filePath , FILE_WRITE)
## filePath.println('date, hour, min, sec, T1, R1, T2, R2, R, G, B, L')
##
textFile_filter_results = open(textFile_filter_results, 'w')
textFile_filter_results.write(
'y_pixel,n_trees,n,n_caught_by_dlc_prob,n_caught_by_slope\n')
x_rows = range(x_1, x_2, stp)
y_rows = range(y_1, y_2, stp)
for xpixel in x_rows:
if 'img' in locals():
img.close()
print('closed img')
Image.MAX_IMAGE_PIXELS = big_img_size
img = PIL.Image.open(path_img_big)
img.load()
path_im_crop_xrow = path_image_tree + str(xpixel) + '/'
if os.path.exists(path_im_crop_xrow):
print(path_im_crop_xrow + ' exists')
else:
print(path_im_crop_xrow + ' made')
os.mkdir(path_im_crop_xrow)
textFile = path_im_crop_xrow + 'predicted_tree_coordinates.txt'
textFile = open(textFile, 'w')
textFile.write(
'raster,x_1,y_1,x_2,y_2,slope,intercept,tree_px,delta_tree,image_px\n'
)
cropped_images = []
for ypixel in y_rows:
left = xpixel
upper = ypixel - 50
right = xpixel + 50
lower = ypixel
path_im_crop = path_im_crop_xrow + str(left) + '_' + str(
upper) + '_' + str(right) + '_' + str(lower) + '.png'
cropped_images.append(path_im_crop)
if os.path.exists(path_im_crop):
print(path_im_crop + ' exists')
else:
im_crop = img.crop((left, upper, right, lower))
im_crop.save(path_im_crop)
im_crop.close()
img.close()
deeplabcut.analyze_time_lapse_frames(config_path,
path_im_crop_xrow,
frametype='.png',
shuffle=2,
save_as_csv=True)
path_dlc_csv = glob.glob(path_im_crop_xrow + '*.csv')
path_dlc_h5 = glob.glob(path_im_crop_xrow +
'*.h5') ## costs a lot of storage space
path_dlc_pickle = glob.glob(
path_im_crop_xrow + '*.pickle') ## costs a lot of storage space
## we want to get rid of both the .pickle and .h5 files since we will not be using them
for h5 in path_dlc_h5:
os.remove(h5)
for pickle in path_dlc_pickle:
os.remove(pickle)
dat = pandas.read_csv(
path_dlc_csv[0]) ## opens up the .csv file created by dlc
dat = dat.iloc[
2:] # gets rid of the first two rows of dat, which are strings and will cause erros
n_caught_by_dlc_prob = 0
n_caught_by_slope = 0
n_trees = 0
n = 0
for row in dat.iterrows():
n = n + 1
crap, data = row
image_dat = data[0]
x_b = float(data[1])
y_b = float(data[2])
p1l = float(data[3])
x_t = float(data[4])
y_t = float(data[5])
p2l = float(data[6])
if (x_t - x_b) != 0:
slope = (y_t - y_b) / (x_t - x_b)
else:
slope = 0
intercept = y_b - slope * x_b
path_img = path_im_crop_xrow + image_dat
if p1l >= 0.9 and p2l >= 0.9:
if slope > -2 and slope < -0.5:
n_trees = n_trees + 1
textFile.write(
os.path.basename(path_img).split('.')[0] + ',')
textFile.write(
str(x_b) + ',' + str(y_b) + ',' + str(x_t) + ',' +
str(y_t) + ',' + str(slope) + ',')
textFile.write(str(intercept) + '\n')
else:
n_caught_by_slope = n_caught_by_slope + 1
#os.remove(path_img)
else:
n_caught_by_dlc_prob = n_caught_by_dlc_prob + 1
os.remove(path_img)
img.close()
textFile.close()
textFile_filter_results.write(str(ypixel) + ',')
textFile_filter_results.write(str(n_trees) + ',')
textFile_filter_results.write(str(n) + ',')
textFile_filter_results.write(str(n_caught_by_dlc_prob) + ',')
textFile_filter_results.write(str(n_caught_by_slope) + '\n')
textFile_filter_results.close()
)
DLC_config = deeplabcut.auxiliaryfunctions.read_plainconfig(posefile)
DLC_config["save_iters"] = 10
DLC_config["display_iters"] = 2
DLC_config["multi_step"] = [[0.001, 10]]
print("CHANGING training parameters to end quickly!")
deeplabcut.auxiliaryfunctions.write_plainconfig(posefile, DLC_config)
print("TRAINING shuffle 2, with smaller allocated memory")
deeplabcut.train_network(path_config_file, shuffle=2, allow_growth=True)
print("ANALYZING some individual frames")
deeplabcut.analyze_time_lapse_frames(
path_config_file,
os.path.join(cfg["project_path"], "labeled-data/reachingvideo1/"),
)
print("Export model...")
deeplabcut.export_model(path_config_file, shuffle=2, make_tar=False)
print("Merging datasets...")
trainIndices, testIndices = deeplabcut.mergeandsplit(
path_config_file, trainindex=0, uniform=True
)
print("Creating two identical splits...")
deeplabcut.create_training_dataset(
path_config_file,
Shuffles=[4, 5],
trainIndices=[trainIndices, trainIndices],
