def test_flownet():
sum_error = 0
sum_px_error = 0
result = open('result.txt', 'wb')
prediction_file = 'flownets-pred-0000000.flo'
img1_list = open('img1_list_test.txt', 'r').readlines()
img2_list = open('img2_list_test.txt', 'r').readlines()
flow_list = open('flo_list_test.txt', 'r').readlines()
length = len(img1_list)
for i in range(length):
img_files = []
img_files.append(img1_list[i].strip())
img_files.append(img2_list[i].strip())
# sanity check
if os.path.exists(prediction_file):
os.remove(prediction_file)
FlowNet.run(this_dir, img_files, './model_simple')
epe = fl.evaluate_flow_file(flow_list[i].strip(), prediction_file)
flow = fl.read_flow(prediction_file)
[height, width, channels] = flow.shape
sum_error += epe
sum_px_error += epe / (height * width)
result.write(str.format("%4d" % i) + ': ' + str(epe) + '\n')
print 'Average Image EPE error: ', sum_error/length
print 'Average Pixel EPE error: ', sum_px_error/length
result.write('Average Image EPE error: ' + str(sum_error / length))
result.write('\n')
result.write('Average Pixel EPE error: ' + str(sum_px_error / length))
result.close()
def test_rain():
rain_image_path = 'haze_rain'
prediction_file = 'flownets-pred-0000000.flo'
left_name_base = 'haze_rain_light/render_haze_left_beta'
right_name_base = 'haze_rain_light/render_haze_right_beta'
flow_file = 'haze_rain_light/flow_left.flo'
result = open('result.txt', 'wb')
sum_error = 0
for beta in range(0, 200, 5):
for contrast in range(120, 201, 5):
img_files = []
left_name = left_name_base + str(beta) + 'contrast' + str(contrast) + '.png'
right_name = right_name_base + str(beta) + 'contrast' + str(contrast) + '.png'
img_files.append(right_name)
img_files.append(left_name)
# sanity check
if os.path.exists(prediction_file):
os.remove(prediction_file)
FlowNet.run(this_dir, img_files, './model_simple')
epe = fl.evaluate_flow_file(flow_file, prediction_file)
flow = fl.read_flow(prediction_file)
flowpic = fl.flow_to_image(flow)
flow_image = Image.fromarray(flowpic)
flow_image.save('beta' + str(beta)+ 'contrast' + str(contrast) + 'flow.png')
sum_error += epe
result.write('beta: ' + str(beta) + ' contrast: ' + str(contrast) + ' epe: ' + str(epe) + '\n')
print 'sum of average end point error: ', sum_error
result.close()
def test_flownet():
sum_error = 0
sum_px_error = 0
result = open('result.txt', 'wb')
prediction_file = 'flownets-pred-0000000.flo'
img1_list = open('img1_list_test.txt', 'r').readlines()
img2_list = open('img2_list_test.txt', 'r').readlines()
flow_list = open('flo_list_test.txt', 'r').readlines()
length = len(img1_list)
for i in range(length):
img_files = []
img_files.append(img1_list[i].strip())
img_files.append(img2_list[i].strip())
# sanity check
if os.path.exists(prediction_file):
os.remove(prediction_file)
FlowNet.run(this_dir, img_files, './model_simple')
epe = fl.evaluate_flow_file(flow_list[i].strip(), prediction_file)
flow = fl.read_flow(prediction_file)
[height, width, channels] = flow.shape
sum_error += epe
sum_px_error += epe / (height * width)
result.write(str.format("%4d" % i) + ': ' + str(epe) + '\n')
print 'Average Image EPE error: ', sum_error / length
print 'Average Pixel EPE error: ', sum_px_error / length
result.write('\n')
result.write('Average Image EPE error: ' + str(sum_error / length))
result.write('Average Pixel EPE error: ' + str(sum_px_error / length))
result.close()
def test_kitti():
kitti_image = '../../../../../../../media/data/LRT_Flow/KITTI/data_scene_flow/training/image_2'
kitti_flow = 'noc'
image_list = os.listdir(kitti_image)
flow_list = os.listdir(kitti_flow)
prediction_file = 'flownets-pred-0000000.flo'
result = open('result.txt', 'wb')
sum_error = 0
# retrieve image and flow information
image_list.sort()
flow_list.sort()
for i in range(0, 308, 2):
# input images and ground truth flow
img_files = []
img_files.append(os.path.join(kitti_image, image_list[i]))
img_files.append(os.path.join(kitti_image, image_list[i + 1]))
ground_truth_file = flow_list[i / 2]
# sanity check
if os.path.exists(prediction_file):
os.remove(prediction_file)
#invoke FlowNet
FlowNet.run(this_dir, img_files, './model_simple')
print i
print img_files[0]
print img_files[1]
print ground_truth_file
#evaluate result
epe = fl.evaluate_flow_file('noc/' + ground_truth_file,
prediction_file)
sum_error += epe
# write to result file
result.write(
str(i) + ':\n' + img_files[0] + '\n' + img_files[1] + '\n' +
ground_truth_file + '\n')
result.write('Average end point error: ' + str(epe) + '\n')
result.write('Total average point error: ' + str(sum_error) + '\n')
print 'sum of average end point error: ', sum_error
result.close()
def test_middlebury():
# setting up
middlebury_image_1 = 'data/test/t0'
middlebury_image_2 = 'data/test/t1'
middlebury_flow = 'data/test/flow'
img1_name = '00000.png'
img2_name = '00000.png'
flow_name = '00000.flo'
prediction_file = 'flownets-pred-0000000.flo'
result_file = 'result.txt'
result = open(result_file, 'wb')
sum_error = 0
# Retrieve image and flow information
folder_names = os.listdir(data / test)
for folder in folder_names:
# input images and ground truth flow
img_files = []
img_files.append(os.path.join(middlebury_image_1, folder, img1_name))
img_files.append(os.path.join(middlebury_image_2, folder, img2_name))
ground_truth_file = os.path.join(middlebury_flow, folder, flow_name)
# sanity check
if os.path.exists(prediction_file):
os.remove(prediction_file)
# invoke FlowNet
FlowNet.run(this_dir, img_files, './model_corr')
# evaluate result
epe = fl.evaluate_flow_file(ground_truth_file, prediction_file)
sum_error += epe
print folder, " average end point error is:", epe
# write to result file
result.write(folder + ':\n' + img_files[0] + '\n' + img_files[1] +
'\n' + ground_truth_file + '\n')
result.write('Average end point error: ' + str(epe) + '\n')
result.write('Total average point error: ' + str(sum_error) + '\n')
print 'sum of average end point error: ', sum_error
result.close()
def test_middlebury():
# setting up
middlebury_image = 'data/other-data'
middlebury_flow = 'data/other-gt-flow'
img1_name = 'frame10_rain.png'
img2_name = 'frame11_rain.png'
flow_name = 'flow10.flo'
prediction_file = 'flownets-pred-0000000.flo'
result_file = 'result.txt'
result = open(result_file , 'wb')
sum_error = 0
# Retrieve image and flow information
folder_names = os.listdir(middlebury_image)
for folder in folder_names:
# input images and ground truth flow
img_files = []
img_files.append(os.path.join(middlebury_image, folder, img1_name))
img_files.append(os.path.join(middlebury_image, folder, img2_name))
ground_truth_file = os.path.join(middlebury_flow, folder, flow_name)
# sanity check
if os.path.exists(prediction_file):
os.remove(prediction_file)
# invoke FlowNet
FlowNet.run(this_dir, img_files, './model_simple')
# evaluate result
epe = fl.evaluate_flow_file(ground_truth_file, prediction_file)
sum_error += epe
print folder, " average end point error is:", epe
# write to result file
result.write(folder + ':\n' + img_files[0] +'\n' + img_files[1] + '\n' + ground_truth_file + '\n')
result.write('Average end point error: ' + str(epe) + '\n')
result.write('Total average point error: ' + str(sum_error) + '\n')
print 'sum of average end point error: ', sum_error
result.close()
def test_kitti():
sum_error = 0
sum_px_error = 0
result = open('result.txt', 'wb')
prediction_file = 'flownets-pred-0000000.flo'
img1_list = open('img1_kitti_test.txt', 'r').readlines()
img2_list = open('img2_kitti_test.txt', 'r').readlines()
flow_list = open('flo_kitti_test.txt', 'r').readlines()
length = len(img1_list)
for i in range(length):
# input images and ground truth flow
img_files = []
img_files.append(img1_list[i].strip())
img_files.append(img2_list[i].strip())
# sanity check
if os.path.exists(prediction_file):
os.remove(prediction_file)
#invoke FlowNet
FlowNet.run(this_dir, img_files, './model_simple')
#evaluate result
flow = fl.read_flow(prediction_file)
gt = kittitool.flow_read(flow_list[i].strip())
epe = fl.evaluate_flow(gt, flow)
sum_error += epe
# write to result file
result.write(
str(i) + ':\n' + img_files[0] + '\n' + img_files[1] + '\n' +
flow_list[i].strip() + '\n')
result.write('Average end point error: ' + str(epe) + '\n')
result.write('Total average point error: ' + str(sum_error) + '\n')
print 'sum of average end point error: ', sum_error
result.close()
def test_kitti():
sum_error = 0
sum_px_error = 0
result = open('result.txt', 'wb')
prediction_file = 'flownets-pred-0000000.flo'
img1_list = open('img1_list_test.txt', 'r').readlines()
img2_list = open('img2_list_test.txt', 'r').readlines()
flow_list = open('flo_list_test.txt', 'r').readlines()
length = len(img1_list)
for i in range(length):
# input images and ground truth flow
img_files = []
img_files.append(img1_list[i].strip())
img_files.append(img2_list[i].strip())
# sanity check
if os.path.exists(prediction_file):
os.remove(prediction_file)
#invoke FlowNet
FlowNet.run(this_dir, img_files, './model_simple')
#evaluate result
flow = fl.read_flow(prediction_file)
gt = kittitool.flow_read(flow_list[i].strip())
epe = fl.evaluate_flow(gt, flow)
sum_error += epe
# write to result file
result.write(str(i) + ':\n' + img_files[0] +'\n' + img_files[1] + '\n' + flow_list[i].strip() + '\n')
result.write('Average end point error: ' + str(epe) + '\n')
result.write('Total average point error: ' + str(sum_error) + '\n')
print 'sum of average end point error: ', sum_error
result.close()
def test_rain():
rain_image_path = 'haze_rain'
prediction_file = 'flownets-pred-0000000.flo'
left_name_base = 'haze_rain_light/render_haze_left_beta'
right_name_base = 'haze_rain_light/render_haze_right_beta'
flow_file = 'haze_rain_light/flow_left.flo'
result = open('result.txt', 'wb')
sum_error = 0
for beta in range(0, 200, 5):
for contrast in range(120, 201, 5):
img_files = []
left_name = left_name_base + str(beta) + 'contrast' + str(
contrast) + '.png'
right_name = right_name_base + str(beta) + 'contrast' + str(
contrast) + '.png'
img_files.append(right_name)
img_files.append(left_name)
# sanity check
if os.path.exists(prediction_file):
os.remove(prediction_file)
FlowNet.run(this_dir, img_files, './model_simple')
epe = fl.evaluate_flow_file(flow_file, prediction_file)
flow = fl.read_flow(prediction_file)
flowpic = fl.flow_to_image(flow)
flow_image = Image.fromarray(flowpic)
flow_image.save('beta' + str(beta) + 'contrast' + str(contrast) +
'flow.png')
sum_error += epe
result.write('beta: ' + str(beta) + ' contrast: ' + str(contrast) +
' epe: ' + str(epe) + '\n')
print 'sum of average end point error: ', sum_error
result.close()
#!/usr/bin/python
import os, sys
from scripts.flownet import FlowNet
my_dir = os.path.dirname(os.path.realpath(__file__))
os.chdir(my_dir + '/..')
if len(sys.argv) -1 != 3:
print("Use this tool to test FlowNet on images\n"
"Usage for single image pair:\n"
" ./demo_flownet.py {S|C} IMAGE1 IMAGE2\n"
"\n"
"Usage for a pair of image lists (must end with .txt):\n"
" ./demo_flownet.py {S|C} LIST1.TXT LIST2.TXT\n")
sys.exit(1)
model_folder = ''
if sys.argv[1].upper() == 'S':
model_folder = './model_simple'
elif sys.argv[1].upper() == 'C':
model_folder = './model_corr'
else:
print("Please specify S for FlowNetSimple or C for FlowNetCorr\n")
sys.exit(1)
img_files = sys.argv[2:]
FlowNet.run(my_dir, img_files, model_folder)
#!/usr/bin/python
import os, sys, shutil
from scripts.flownet import FlowNet
# get root directory
my_dir = os.path.dirname(os.path.realpath(__file__))
os.chdir(my_dir + '/..')
# configuration
model_folder = './model_simple'
img_files = sorted(os.listdir('flownet/data/shaman_3/'))
# Running flow and store them up
for i in range(0, len(img_files)-1):
filename_1 = os.path.join('data/shaman_3/', img_files[i])
filename_2 = os.path.join('data/shaman_3/', img_files[i+1])
FlowNet.run(my_dir, [filename_1, filename_2], model_folder)
src = os.path.join(my_dir, 'flownets-pred-0000000.flo')
dst = os.path.join(my_dir, 'out/', filename_1+'.flo')
shutil.move(src,dst)
if ret is True:
indFrame = indFrame + 1
if ((indFrame - 1) % step) == 0 and indFlowMap < numFlowMap:
imgDisplay = np.hstack((prvs, next))
# save the frames into png files for FlowNet to read
# TODO: stupid but is the easiest way without reconfigure
# the FlowNet and possible re-train the model
cv2.imwrite('data/frame1.png', prvs)
cv2.imwrite('data/frame2.png', next)
# compute the optical flow from two adjacent frames
# the FlowNet will save a .flo file (input is a frame?)
FlowNet.run(prvs)
# read the .flo file
nameFlow = 'flownetc-pred-0000000.flo'
# temporary output image in the Middlebury color style
nameOutTemp = 'outTemp.ppm'
# convert the flow file to the color image file
ColorFlow.flow2color(nameOutTemp, nameFlow)
# read the temporary output image (.ppm)
img = cv2.imread(nameOutTemp)
imgDisplay = np.hstack((imgDisplay, img))
out.write(img)
#!/usr/bin/python
import os, sys
from scripts.flownet import FlowNet
my_dir = os.path.dirname(os.path.realpath(__file__))
os.chdir(my_dir + '/..')
if len(sys.argv) - 1 <= 1:
print("Use this tool to train FlowNet using the train.prototxt\n"
"Usage for FlowNetS or FlowNetC:\n"
" ./train_flownet.py {S|C}\n")
sys.exit(1)
model_folder = ''
if sys.argv[1].upper() == 'S':
model_folder = './model_simple'
elif sys.argv[1].upper() == 'C':
model_folder = './model_corr'
else:
print("Please specify S for FlowNetSimple or C for FlowNetCorr\n")
sys.exit(1)
FlowNet.train(my_dir, model_folder, sys.argv[2:])
#!/usr/bin/python
import os, sys
from scripts.flownet import FlowNet
my_dir = os.path.dirname(os.path.realpath(__file__))
os.chdir(my_dir + '/..')
if len(sys.argv) - 1 != 3:
print("Use this tool to test FlowNet on images\n"
"Usage for single image pair:\n"
" ./demo_flownet.py {S|C} IMAGE1 IMAGE2\n"
"\n"
"Usage for a pair of image lists (must end with .txt):\n"
" ./demo_flownet.py {S|C} LIST1.TXT LIST2.TXT\n")
sys.exit(1)
model_folder = ''
if sys.argv[1].upper() == 'S':
model_folder = './model_simple'
elif sys.argv[1].upper() == 'C':
model_folder = './model_corr'
else:
print("Please specify S for FlowNetSimple or C for FlowNetCorr\n")
sys.exit(1)
img_files = sys.argv[2:]
FlowNet.run(my_dir, img_files, model_folder)
if ret is True:
indFrame = indFrame + 1
if ((indFrame - 1) % step) == 0 and indFlowMap < numFlowMap:
imgDisplay = np.hstack((prvs, next))
# save the frames into png files for FlowNet to read
# TODO: stupid but is the easiest way without reconfigure
# the FlowNet and possible re-train the model
cv2.imwrite('data/frame1.png', prvs)
cv2.imwrite('data/frame2.png', next)
# compute the optical flow from two adjacent frames
FlowNet.run(prvs) # the FlowNet will save a .flo file
# read the .flo file
fileName = 'flownetc-pred-0000000.flo'
flowMapSize = np.fromfile(fileName, np.float32, count=1)
if flowMapSize != 202021.25:
print 'Dimension incorrect. Invalid .flo file'
else:
data = np.fromfile(fileName, np.float32,
count=2 * width * height)
flow = np.resize(data, (height, width, 2))
# prunt the flow value if it's weirdly large
for index, x in np.ndenumerate(flow):
if x > 500:
if ret is True:
indFrame = indFrame + 1
if ((indFrame - 1) %
step) == 0 and indFlowMap < numFlowMap:
imgDisplay = np.hstack((prvs, next))
# save the frames into png files for FlowNet to read
# TODO: stupid but is the easiest way without reconfigure
# the FlowNet and possible re-train the model
cv2.imwrite('data/frame1.png', prvs)
cv2.imwrite('data/frame2.png', next)
# compute the optical flow from two adjacent frames
FlowNet.run(prvs) # the FlowNet will save a .flo file
# read the .flo file
nameFlow = 'flownetc-pred-0000000.flo'
flowMapSize = np.fromfile(nameFlow,
np.float32,
count=1)
if flowMapSize != 202021.25:
print 'Dimension incorrect. Invalid .flo file'
else:
data = np.fromfile(nameFlow,
np.float32,
count=2 * width * height)
flow = np.resize(data, (height, width, 2))
#!/usr/bin/python
import os, sys
from scripts.flownet import FlowNet
my_dir = os.path.dirname(os.path.realpath(__file__))
os.chdir(my_dir + '/..')
if len(sys.argv)-1 < 1:
print("Use this tool to train FlowNet using the train.prototxt\n"
"Usage for FlowNetS or FlowNetC:\n"
" ./train_flownet.py {S|C}\n")
sys.exit(1)
model_folder = ''
if sys.argv[1].upper() == 'S':
model_folder = './model_simple'
elif sys.argv[1].upper() == 'C':
model_folder = './model_corr'
else:
print("Please specify S for FlowNetSimple or C for FlowNetCorr\n")
sys.exit(1)
FlowNet.train(my_dir, model_folder, sys.argv[2:])
