def Read_Input_Images(inputDir, listOfIgnoredSamples, dB, resizedFlag, table,
workplace, spatial_size, channel, objective_flag):
r = w = spatial_size
SubperdB = []
# cross-checking parameter
subperdb_id = []
for sub in sorted([infile for infile in os.listdir(inputDir)]):
VidperSub = []
vid_id = np.empty([0])
for vid in sorted([inrfile for inrfile in os.listdir(inputDir + sub)]):
path = inputDir + sub + '/' + vid + '/' # image loading path
if path in listOfIgnoredSamples:
continue
imgList = readinput(path)
numFrame = len(imgList)
if resizedFlag == 1:
col = w
row = r
else:
img = cv2.imread(imgList[0])
[row, col, _l] = img.shape
## read the label for each input video
collectinglabel(table, sub[3:], vid, workplace + 'Classification/',
dB, objective_flag)
for var in range(numFrame):
img = cv2.imread(imgList[var])
[_, _, dim] = img.shape
if channel == 1:
img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
if resizedFlag == 1:
img = cv2.resize(img, (col, row))
if var == 0:
FrameperVid = img.flatten()
else:
FrameperVid = np.vstack((FrameperVid, img.flatten()))
vid_id = np.append(vid_id, imgList[var]) # <--cross-check
VidperSub.append(FrameperVid)
subperdb_id.append(vid_id) # <--cross-check
SubperdB.append(VidperSub)
return SubperdB
def augmentation_casme(db_images, outputDir, numSamples, table, resizedFlag, r, w):
for emotion in ['positive', 'negative', 'surprise', 'others']:
table_emotion = pd.DataFrame(data=table[0:,0:],columns=['sub','id','emotion'])
table_emotion = table_emotion[table_emotion['emotion']==emotion]
for i in range(numSamples):
print(emotion+"_"+str(i))
# first we ensure that every original video is processed, then we start sampling randomly until we have enough
if i <= (table_emotion.shape[0]-1):
random_pick = table_emotion.iloc[[i]] # not so random
print("processing original video #" + str(i) + " emotion=" + emotion)
else:
random_pick = table_emotion.sample(n=1) # very random
print("processing augmented video #" + str(i) + "emotion" + emotion)
path = db_images+"sub"+str(random_pick['sub'].iloc[0])+"/"+str(random_pick['id'].iloc[0])+"/"
imgList = readinput(path)
numFrame = len(imgList)
if resizedFlag == 1:
col = w
row = r
else:
img = cv2.imread(imgList[0])
[row,col,_l] = img.shape
cropBool = 1#randint(0,1)
cropStyle = randint(1,8)
flipBool = randint(0,1)
rotBool = 0#randint(0,1)
rotAngle = 0#randint(-8,8)
for var in range(numFrame):
img = cv2.imread(imgList[var])
[_,_,dim] = img.shape
if resizedFlag == 1:
img = cv2.resize(img, (col,row))
if i > (table_emotion.shape[0]-1):
img = augment_image(img,cropBool,cropStyle,flipBool,rotBool, rotAngle)
writeFolder = outputDir+"sub"+str(random_pick['sub'].iloc[0])+"/"+str(random_pick['id'].iloc[0])+"."+str(i)+"/"
outputPath = writeFolder + imgList[var].split('/')[-1]
if not os.path.exists(writeFolder):
os.makedirs(writeFolder)
cv2.imwrite(outputPath, img)
def Read_Input_Images_Apex(inputDir, listOfIgnoredSamples, dB, resizedFlag,
table, workplace, spatial_size, channel,
objective_flag):
r = w = spatial_size
SubperdB = []
# cross-checking parameter
subperdb_id = []
for sub in sorted([infile for infile in os.listdir(inputDir)]):
VidperSub = []
for vid in sorted([inrfile for inrfile in os.listdir(inputDir + sub)]):
path = inputDir + sub + '/' + vid + '/' # image loading path
if path in listOfIgnoredSamples:
print(path)
continue
imgList = readinput(path)
collectinglabel(table, sub[3:],
vid.split(".")[0], workplace + 'Classification/',
dB, objective_flag)
table_df = pd.DataFrame(table)
apexInd = table_df[table_df[0] == sub[-2:]]
apexInd = int(apexInd[apexInd[1] == vid][3])
imgON = cv2.imread(imgList[0])
imgApex = cv2.imread(imgList[apexInd])
if channel == 1:
imgON = cv2.cvtColor(imgON, cv2.COLOR_BGR2GRAY)
imgApex = cv2.cvtColor(imgApex, cv2.COLOR_BGR2GRAY)
if resizedFlag == 1:
imgON = cv2.resize(imgON, (spatial_size, spatial_size))
imgApex = cv2.resize(imgApex, (spatial_size, spatial_size))
FrameperVid = imgON.flatten()
FrameperVid = np.vstack((FrameperVid, imgApex.flatten()))
VidperSub.append(FrameperVid)
SubperdB.append(VidperSub)
return SubperdB
def augmenting2():
new_image = 0
for sub in sorted([infile for infile in os.listdir(img_path)]):
for vid in sorted([inrfile for inrfile in os.listdir(img_path + sub)]):
augmentations = np.empty([0])
path = img_path + sub + '/' + vid + '/'
folder_path = img_path + sub + '/'
if path in listOfIgnoredSamples:
continue
# print(path)
imgList = readinput(path, dB)
numFrame = len(imgList)
# print(numFrame)
# print("imglist: %i" % imgList)
for var in range(numFrame):
augmentations = pixel_removal(2, imgList[var])
# print("augmentations: %i" % len(augmentations))
# print("var: %i" % var)
counter = 0
while counter < len(augmentations):
# # split augmented data into different folders
folder_name = vid + "_" + str(counter)
folder_name = folder_path + folder_name + "/"
# print(folder_name)
if not os.path.exists(folder_name):
os.makedirs(folder_name)
if var < 9:
filename = folder_name + "00" + str(var + 1) + ".jpg"
else:
filename = folder_name + "0" + str(var + 1) + ".jpg"
os.chdir(folder_name)
cv2.imwrite(filename, augmentations[counter])
os.chdir(path)
counter += 1
print("Written to: " + filename)
new_image += 1
print("Total new images: " + str(new_image))
def cube_maker():
counter = 1
for sub in sorted([infile for infile in os.listdir(img_path)]):
for vid in sorted([inrfile for inrfile in os.listdir(img_path + sub)]):
path = img_path + sub + '/' + vid + '/'
if path in listOfIgnoredSamples:
continue
print(path)
imgList = readinput(path, dB)
numFrame = len(imgList)
for var in range(numFrame):
img = cv2.imread(imgList[var])
img = cv2.resize(img, (256, 256))
cv2.imwrite(imgList[var], img)
counter += 1
print(str(counter) + "/5659")
else:
print("NOT in the selection.")
######### Reading in the input images ########
SubperdB = []
for sub in sorted([infile for infile in os.listdir(inputDir)]):
VidperSub = []
for vid in sorted([inrfile for inrfile in os.listdir(inputDir + sub)]):
path = inputDir + sub + '/' + vid + '/'
if path in listOfIgnoredSamples:
continue
imgList = readinput(path, dB)
numFrame = len(imgList)
if resizedFlag == 1:
col = w
row = r
else:
img = cv2.imread(imgList[0])
[row, col, _l] = img.shape
## read the label for each input video
collectinglabel(table, sub[3:], vid, workplace + 'Classification/', dB)
for var in range(numFrame):
def Read_Input_Images_SAMM_CASME(inputDir, filteredSamples, ignoredSamples, dB,
resizedFlag, table, workplace, spatial_size,
channel, objective_flag):
# r=224; w=224
r = w = spatial_size
SubperdB = []
# cross-checking parameter
subperdb_id = []
for sub in sorted([infile for infile in os.listdir(inputDir)]):
VidperSub = []
vid_id = np.empty([0])
for vid in sorted([inrfile for inrfile in os.listdir(inputDir + sub)]):
path = inputDir + sub + '/' + vid + '/'
# print(len(filteredSamples))
# print(filteredSamples)
# print("bohaha")
# filtered samples are samples needed
if path not in filteredSamples:
# print(path)
continue
# print(dB)
# print(path)
imgList = readinput(path)
numFrame = len(imgList)
# print(numFrame)
if resizedFlag == 1:
col = w
row = r
else:
img = cv2.imread(imgList[0])
[row, col, _l] = img.shape
## ##read the label for each input video
# print(sub[3:])
collectinglabel(table, sub, vid, workplace + 'Classification/', dB,
objective_flag)
for var in range(numFrame):
img = cv2.imread(imgList[var])
[_, _, dim] = img.shape
if channel == 1:
img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
if resizedFlag == 1:
#in resize function, [col,row]
img = cv2.resize(img, (col, row))
if var == 0:
FrameperVid = img.flatten()
else:
FrameperVid = np.vstack((FrameperVid, img.flatten()))
vid_id = np.append(vid_id, imgList[var]) # <--cross-check
VidperSub.append(FrameperVid)
subperdb_id.append(vid_id) # <--cross-check
# print(subperdb_id)
# if len(VidperSub) > 0:
# print(len(VidperSub))
SubperdB.append(VidperSub)
# return SubperdB, vid_id, subperdb_id
return SubperdB
def augmenting():
new_image = 0
for sub in sorted([infile for infile in os.listdir(img_path)]):
for vid in sorted([inrfile for inrfile in os.listdir(img_path + sub)]):
augmentations = np.empty([0])
path = img_path + sub + '/' + vid + '/'
folder_path = img_path + sub + '/'
if path in listOfIgnoredSamples:
continue
# print(path)
imgList = readinput(path, dB)
numFrame = len(imgList)
# print(numFrame)
# hardcoded
number_of_ops = 13
folder_num = 0
for var in range(numFrame):
img = cv2.imread(imgList[var])
rotated_img = rotation(5, img)
rotated_img_2 = rotation(-5, img)
rotated_img_3 = rotation(10, img)
rotated_img_4 = rotation(-10, img)
scaled_img0_9 = scaling(0.9, img)
scaled_img1_0 = scaling(1.0, img)
scaled_img1_1 = scaling(1.1, img)
translation_1 = translation(-2, -2, img)
translation_2 = translation(-2, 2, img)
translation_3 = translation(2, -2, img)
translation_4 = translation(2, 2, img)
horizontal, vertical = flip(img)
augmentations = np.append(augmentations, [
rotated_img, rotated_img_2, rotated_img_3, rotated_img_4,
scaled_img0_9, scaled_img1_0, scaled_img1_1, translation_1,
translation_2, translation_3, translation_4, horizontal,
vertical
])
counter = 0
seq_var = -9
count_seq = 0
filename_count = 1
while counter < len(augmentations):
# split augmented data into different folders
folder_name = vid + "_" + str(folder_num)
folder_name = folder_path + folder_name + "/"
# print(folder_name)
if not os.path.exists(folder_name):
os.makedirs(folder_name)
# filename = path + "0" + str( numFrame + counter + 1 ) + ".jpg"
# filename = folder_name + "0" + str( numFrame + counter + 1 ) + ".jpg"
if filename_count < 10:
filename = folder_name + "00" + str(
filename_count) + ".jpg"
else:
filename = folder_name + "0" + str(filename_count) + ".jpg"
# print(str(seq_var))
print(filename)
os.chdir(folder_name)
cv2.imwrite(filename, augmentations[seq_var + 13])
os.chdir(path)
counter += 1
new_image += 1
seq_var += 13
filename_count += 1
if counter % 10 == 0 and counter > 0:
count_seq += 1
seq_var = -13 + count_seq
folder_num += 1
filename_count = 1
# print("Saved to " + path)
# seq_var += 1
print("Total new images: " + str(new_image))