def parse_args():
parser = argparse.ArgumentParser()
parser.add_argument('--exp-name', type=str, default='C20_S1_0.1_CS')
parser.add_argument('--split', type=str, default='val')
parser.add_argument('--save-each-confmat', type=bool, default=True)
# parser.add_argument('--dir-struct', type=str, default='deeplab')
# parser.add_argument('--class-subset', type=int, nargs='+', default=None)
parser.add_argument('--dir-struct', type=str, default='mtpt')
parser.add_argument('--class-subset',
type=int,
nargs='+',
default=[0, 1, 2, 3, 4, 5, 7, 8, 11, 13, 19])
# parser.add_argument('--predict-list', type=str,
# help='list of predictions')
# parser.add_argument('--target-list', type=str,
# help='list of targets')
# parser.add_argument('--output-dir', type=str,
# help='directory stores the output')
##
opts = parser.parse_args()
opts.target_list = sorted(
glob.glob(join(data_dir,
'CityScapes/ReOrg/SegColor20-1024/val/*.png')))
if opts.dir_struct == 'deeplab':
predict_dir = join(data_dir, 'Exp/DeepLabCityscapes', opts.exp_name,
'predict')
opts.predict_list = sorted(
glob.glob(join(predict_dir, opts.split, '*.png')))
record_dir = mkdir2(join(data_dir, 'Exp/DeepLabCityscapes/record'))
elif opts.dir_struct == 'mtpt':
predict_dir = join(data_dir, 'Exp/Cityscapes/mtpt/output',
opts.exp_name)
opts.predict_list = sorted(
glob.glob(join(predict_dir, 'pred', '*.png')))
record_dir = mkdir2(join(data_dir, 'Exp/Cityscapes/record'))
else:
raise ValueError('Undefined directory structure')
opts.outfile_record_entry = join(record_dir, opts.exp_name + '.csv')
opts.outfile_summary = join(predict_dir,
'eval-' + opts.split + '-summary.txt')
opts.outfile_per_class = join(
predict_dir, 'eval-' + opts.split + '-summary-per-class.txt')
opts.outfile_per_inst = join(predict_dir,
'eval-' + opts.split + '-per-inst.txt')
opts.outfile_per_inst_per_class = join(
predict_dir, 'eval-' + opts.split + '-per-inst-per-class.txt')
if opts.save_each_confmat:
opts.confmat_dir = mkdir2(join(predict_dir, 'confmat', opts.split))
opts.confmat_total = join(predict_dir, 'confmat', opts.split + '.txt')
return opts
car_mask = (img == 10).astype(np.uint8)
# This point is always the ego-vehicle thus not other car masks
seed = (w // 2, h - 1)
dummy_mask = np.zeros((h + 2, w + 2), np.uint8)
cv2.floodFill(car_mask, dummy_mask, seed, 255)
new_mask = car_mask != 255
img[new_mask] = 10
return img
single_mode = 0
if single_mode:
## single mode
in_dir = join(data_dir, 'Exp/CARLA_gen20_town2/e000003/SegRaw')
out_dir = mkdir2(join(data_dir, 'Exp/CARLA_gen20_town2/e000003/Seg'))
mask_file = join(data_dir, 'Exp/CARLA_gen20_town2/ego-vehicle.png')
mask = np.array(Image.open(mask_file), dtype=bool)
files = glob.glob(join(in_dir, '*.png'))
for f in files:
img = np.array(Image.open(f))
img = img[:, :, 0]
img[mask] = 13
img = proc_sky(img)
img = proc_car(img)
img = Image.fromarray(c2trainid(img).astype(np.uint8))
img.putpalette(cm)
img.save(join(out_dir, os.path.basename(f)))
else:
# parser.add_argument('--dir-struct', type=str, default='deeplab')
# parser.add_argument('--class-subset', type=int, nargs='+', default=None)
parser.add_argument('--dir-struct', type=str, default='mtpt')
parser.add_argument('--class-subset',
type=int,
nargs='+',
default=[0, 1, 2, 3, 4, 5, 7, 8, 11, 13, 19])
opts = parser.parse_args()
exp_name = opts.exp_name
if opts.dir_struct == 'deeplab':
exp_dir = join(data_dir, 'Exp/DeepLabCityscapes', exp_name)
predict_dir = join(exp_dir, 'predict')
out_dir = mkdir2(join(exp_dir, 'vis'))
out_name = 'index.html'
reorg_dir = join(out_dir, '../../../../Cityscapes/ReOrg')
inputs = join(reorg_dir, 'Image-512-JPG/val/*.jpg')
outputs = join(out_dir, '../predict/val/*.png')
targets = join(reorg_dir, 'SegColor20-1024/val/*.png')
targets_org = join(reorg_dir, 'SegColor35-1024/val/*.png')
external_path = 'http://mt1080.pc.cs.cmu.edu:21368/Exp/DeepLabCityscapes/' + exp_name + '/vis/' + out_name
elif opts.dir_struct == 'mtpt':
predict_dir = join(data_dir, 'Exp/Cityscapes/mtpt', 'output', exp_name)
out_dir = predict_dir
out_name = 'advanced.html'
reorg_dir = join(out_dir, '../../../../../Cityscapes/ReOrg')
inputs = join(reorg_dir, 'Image-512-JPG/val/*.jpg')
outputs = join(out_dir, 'pred/*.png')
targets = join(reorg_dir, 'SegColor20-1024/val/*.png')
import sys
sys.path.insert(0, '../util')
from platform_config import data_dir, mkdir2
import os, glob, random, math
from os.path import join
from shutil import copy2
import numpy as np
# for converting to indexed images
from PIL import Image
from cityscapes import c2clabelfromrgb, cm_train
build_dir = mkdir2(join(data_dir,
'Exp/C20_S1')) # gen 18, weather 2, subsampling step 1
train_source_dir = join(data_dir, 'Exp/CARLA_gen20')
test_source_dir = join(data_dir, 'Exp/CARLA_gen20_town2')
rgb_dir = mkdir2(join(build_dir, 'RGB-1024'))
seg_dir = mkdir2(join(build_dir, 'Seg-1024'))
n_frame = 10000
train_seqs = glob.glob(join(train_source_dir, 'e*'))
train_seqs = sorted(list(filter(lambda s: os.path.isdir(s), train_seqs)))
test_seqs = glob.glob(join(test_source_dir, 'e*'))
test_seqs = sorted(list(filter(lambda s: os.path.isdir(s), test_seqs)))
n_val = 2
n_train = len(train_seqs) - n_val
## Setting up platform dependent stuffs import sys; sys.path.insert(0, '../../util') from platform_config import data_dir, mkdir2 from os.path import join, basename from html4vision import Col, imagetable out_dir = mkdir2(join(data_dir, 'Exp/C2C/Visual')) exp_name = 'test6' # exp_name = basename(__file__)[:-3] seq_dir = join(data_dir, 'Exp/CARLA_gen19/e000001') exp_dir = join(data_dir, 'Exp/C2C', exp_name) seg_colors = join(seq_dir, 'SegColor/*.png') source_imgs = join(seq_dir, 'RGB/*.png') matched = join(exp_dir, 'SegMatch-256-Vis/*.jpg') canvas = join(exp_dir, 'Canvas-512-Vis/*.jpg') refined = join(exp_dir, 'Refined-512/*.png') road1_source = join(exp_dir, 'SegMatch-256-VisEx/*_road1.jpg') car1_source = join(exp_dir, 'SegMatch-256-VisEx/*_car1.jpg') car2_source = join(exp_dir, 'SegMatch-256-VisEx/*_car2.jpg') car3_source = join(exp_dir, 'SegMatch-256-VisEx/*_car3.jpg') building1_source = join(exp_dir, 'SegMatch-256-VisEx/*_building1.jpg') sel = None # all
import sys
sys.path.insert(0, '../util')
from platform_config import data_dir, mkdir2
from glob import glob
from os.path import join
from shutil import copy2
from PIL import Image
split = 'train'
in_dir = join(r'D:\c2c2', split, 'images')
out_dir128 = mkdir2(join(data_dir, 'CarlaGen/C18_W2_S1/CycleGAN-128', split))
out_dir1024 = mkdir2(join(data_dir, 'CarlaGen/C18_W2_S1/CycleGAN-1024', split))
file_list = sorted(glob(join(in_dir, '*_fake_B.png')))
n = len(file_list)
for i, f in enumerate(file_list):
print('Processing %d/%d - %s' % (i, n, f))
copy2(f, join(out_dir128, '%08d.png' % (i + 1)))
img = Image.open(f)
img = img.resize((2048, 1024), Image.BICUBIC)
img.save(join(out_dir1024, '%08d.png' % (i + 1)))