Spaces:
Runtime error
Runtime error
import argparse | |
import os | |
import platform | |
import struct | |
import subprocess | |
import time | |
from typing import List | |
import cv2 | |
import numpy as np | |
import torch.multiprocessing as mp | |
from numba import njit | |
import sys | |
sys.path.append("./src/ebsynth/") | |
import blender.histogram_blend as histogram_blend | |
from blender.guide import (BaseGuide, ColorGuide, EdgeGuide, PositionalGuide, | |
TemporalGuide) | |
from blender.poisson_fusion import poisson_fusion | |
from blender.video_sequence import VideoSequence | |
from flow.flow_utils import flow_calc | |
from src.video_util import frame_to_video | |
OPEN_EBSYNTH_LOG = False | |
MAX_PROCESS = 8 | |
os_str = platform.system() | |
if os_str == 'Windows': | |
ebsynth_bin = '.\\src\\ebsynth\\deps\\ebsynth\\bin\\ebsynth.exe' | |
elif os_str == 'Linux': | |
ebsynth_bin = './src/ebsynth/deps/ebsynth/bin/ebsynth' | |
elif os_str == 'Darwin': | |
ebsynth_bin = './src/ebsynth/deps/ebsynth/bin/ebsynth.app' | |
else: | |
print('Cannot recognize OS. Run Ebsynth failed.') | |
exit(0) | |
def g_error_mask_loop(H, W, dist1, dist2, output, weight1, weight2): | |
for i in range(H): | |
for j in range(W): | |
if weight1 * dist1[i, j] < weight2 * dist2[i, j]: | |
output[i, j] = 0 | |
else: | |
output[i, j] = 1 | |
if weight1 == 0: | |
output[i, j] = 0 | |
elif weight2 == 0: | |
output[i, j] = 1 | |
def g_error_mask(dist1, dist2, weight1=1, weight2=1): | |
H, W = dist1.shape | |
output = np.empty_like(dist1, dtype=np.byte) | |
g_error_mask_loop(H, W, dist1, dist2, output, weight1, weight2) | |
return output | |
def create_sequence(base_dir, key_ind, key_dir): | |
sequence = VideoSequence(base_dir, key_ind, 'video', key_dir, | |
'tmp', '%04d.png', '%04d.png') | |
return sequence | |
def process_one_sequence(i, video_sequence: VideoSequence): | |
interval = video_sequence.interval(i) | |
for is_forward in [True, False]: | |
input_seq = video_sequence.get_input_sequence(i, is_forward) | |
output_seq = video_sequence.get_output_sequence(i, is_forward) | |
flow_seq = video_sequence.get_flow_sequence(i, is_forward) | |
key_img_id = i if is_forward else i + 1 | |
key_img = video_sequence.get_key_img(key_img_id) | |
for j in range(interval - 1): | |
i1 = cv2.imread(input_seq[j]) | |
i2 = cv2.imread(input_seq[j + 1]) | |
flow_calc.get_flow(i1, i2, flow_seq[j]) | |
guides: List[BaseGuide] = [ | |
ColorGuide(input_seq), | |
EdgeGuide(input_seq, | |
video_sequence.get_edge_sequence(i, is_forward)), | |
TemporalGuide(key_img, output_seq, flow_seq, | |
video_sequence.get_temporal_sequence(i, is_forward)), | |
PositionalGuide(flow_seq, | |
video_sequence.get_pos_sequence(i, is_forward)) | |
] | |
weights = [6, 0.5, 0.5, 2] | |
for j in range(interval): | |
# key frame | |
if j == 0: | |
img = cv2.imread(key_img) | |
cv2.imwrite(output_seq[0], img) | |
else: | |
cmd = f'{ebsynth_bin} -style {os.path.abspath(key_img)}' | |
for g, w in zip(guides, weights): | |
cmd += ' ' + g.get_cmd(j, w) | |
cmd += (f' -output {os.path.abspath(output_seq[j])}' | |
' -searchvoteiters 12 -patchmatchiters 6') | |
if OPEN_EBSYNTH_LOG: | |
print(cmd) | |
subprocess.run(cmd, | |
shell=True, | |
capture_output=not OPEN_EBSYNTH_LOG) | |
def process_sequences(i_arr, video_sequence: VideoSequence): | |
for i in i_arr: | |
process_one_sequence(i, video_sequence) | |
def run_ebsynth(video_sequence: VideoSequence): | |
beg = time.time() | |
processes = [] | |
mp.set_start_method('spawn') | |
n_process = min(MAX_PROCESS, video_sequence.n_seq) | |
cnt = video_sequence.n_seq // n_process | |
remainder = video_sequence.n_seq % n_process | |
prev_idx = 0 | |
for i in range(n_process): | |
task_cnt = cnt + 1 if i < remainder else cnt | |
i_arr = list(range(prev_idx, prev_idx + task_cnt)) | |
prev_idx += task_cnt | |
p = mp.Process(target=process_sequences, args=(i_arr, video_sequence)) | |
p.start() | |
processes.append(p) | |
for p in processes: | |
p.join() | |
end = time.time() | |
print(f'ebsynth: {end-beg}') | |
def assemble_min_error_img_loop(H, W, a, b, error_mask, out): | |
for i in range(H): | |
for j in range(W): | |
if error_mask[i, j] == 0: | |
out[i, j] = a[i, j] | |
else: | |
out[i, j] = b[i, j] | |
def assemble_min_error_img(a, b, error_mask): | |
H, W = a.shape[0:2] | |
out = np.empty_like(a) | |
assemble_min_error_img_loop(H, W, a, b, error_mask, out) | |
return out | |
def load_error(bin_path, img_shape): | |
img_size = img_shape[0] * img_shape[1] | |
with open(bin_path, 'rb') as fp: | |
bytes = fp.read() | |
read_size = struct.unpack('q', bytes[:8]) | |
assert read_size[0] == img_size | |
float_res = struct.unpack('f' * img_size, bytes[8:]) | |
res = np.array(float_res, | |
dtype=np.float32).reshape(img_shape[0], img_shape[1]) | |
return res | |
def process_seq(video_sequence: VideoSequence, | |
i, | |
blend_histogram=True, | |
blend_gradient=True): | |
key1_img = cv2.imread(video_sequence.get_key_img(i)) | |
img_shape = key1_img.shape | |
interval = video_sequence.interval(i) | |
beg_id = video_sequence.get_sequence_beg_id(i) | |
oas = video_sequence.get_output_sequence(i) | |
obs = video_sequence.get_output_sequence(i, False) | |
binas = [x.replace('jpg', 'bin') for x in oas] | |
binbs = [x.replace('jpg', 'bin') for x in obs] | |
obs = [obs[0]] + list(reversed(obs[1:])) | |
inputs = video_sequence.get_input_sequence(i) | |
oas = [cv2.imread(x) for x in oas] | |
obs = [cv2.imread(x) for x in obs] | |
inputs = [cv2.imread(x) for x in inputs] | |
flow_seq = video_sequence.get_flow_sequence(i) | |
dist1s = [] | |
dist2s = [] | |
for i in range(interval - 1): | |
bin_a = binas[i + 1] | |
bin_b = binbs[i + 1] | |
dist1s.append(load_error(bin_a, img_shape)) | |
dist2s.append(load_error(bin_b, img_shape)) | |
lb = 0 | |
ub = 1 | |
beg = time.time() | |
p_mask = None | |
# write key img | |
blend_out_path = video_sequence.get_blending_img(beg_id) | |
cv2.imwrite(blend_out_path, key1_img) | |
for i in range(interval - 1): | |
c_id = beg_id + i + 1 | |
blend_out_path = video_sequence.get_blending_img(c_id) | |
dist1 = dist1s[i] | |
dist2 = dist2s[i] | |
oa = oas[i + 1] | |
ob = obs[i + 1] | |
weight1 = i / (interval - 1) * (ub - lb) + lb | |
weight2 = 1 - weight1 | |
mask = g_error_mask(dist1, dist2, weight1, weight2) | |
if p_mask is not None: | |
flow_path = flow_seq[i] | |
flow = flow_calc.get_flow(inputs[i], inputs[i + 1], flow_path) | |
p_mask = flow_calc.warp(p_mask, flow, 'nearest') | |
mask = p_mask | mask | |
p_mask = mask | |
# Save tmp mask | |
# out_mask = np.expand_dims(mask, 2) | |
# cv2.imwrite(f'mask/mask_{c_id:04d}.jpg', out_mask * 255) | |
min_error_img = assemble_min_error_img(oa, ob, mask) | |
if blend_histogram: | |
hb_res = histogram_blend.blend(oa, ob, min_error_img, | |
(1 - weight1), (1 - weight2)) | |
else: | |
# hb_res = min_error_img | |
tmpa = oa.astype(np.float32) | |
tmpb = ob.astype(np.float32) | |
hb_res = (1 - weight1) * tmpa + (1 - weight2) * tmpb | |
# cv2.imwrite(blend_out_path, hb_res) | |
# gradient blend | |
if blend_gradient: | |
res = poisson_fusion(hb_res, oa, ob, mask) | |
else: | |
res = hb_res | |
cv2.imwrite(blend_out_path, res) | |
end = time.time() | |
print('others:', end - beg) | |
def main(args): | |
global MAX_PROCESS | |
MAX_PROCESS = args.n_proc | |
video_sequence = create_sequence(f'{args.name}', args.key_ind, args.key) | |
if not args.ne: | |
run_ebsynth(video_sequence) | |
blend_histogram = True | |
blend_gradient = args.ps | |
for i in range(video_sequence.n_seq): | |
process_seq(video_sequence, i, blend_histogram, blend_gradient) | |
if args.output: | |
frame_to_video(args.output, video_sequence.blending_dir, args.fps, | |
False) | |
if not args.tmp: | |
video_sequence.remove_out_and_tmp() | |
if __name__ == '__main__': | |
parser = argparse.ArgumentParser() | |
parser.add_argument('name', type=str, help='Path to input video') | |
parser.add_argument('--output', | |
type=str, | |
default=None, | |
help='Path to output video') | |
parser.add_argument('--fps', | |
type=float, | |
default=30, | |
help='The FPS of output video') | |
parser.add_argument("--key_ind", type=int, nargs='+', default=[1], help="key frame index") | |
parser.add_argument('--key', | |
type=str, | |
default='keys0', | |
help='The subfolder name of stylized key frames') | |
parser.add_argument('--n_proc', | |
type=int, | |
default=8, | |
help='The max process count') | |
parser.add_argument('-ps', | |
action='store_true', | |
help='Use poisson gradient blending') | |
parser.add_argument( | |
'-ne', | |
action='store_true', | |
help='Do not run ebsynth (use previous ebsynth output)') | |
parser.add_argument('-tmp', | |
action='store_true', | |
help='Keep temporary output') | |
args = parser.parse_args() | |
main(args) | |