-
Notifications
You must be signed in to change notification settings - Fork 8
/
run.py
138 lines (111 loc) · 5.1 KB
/
run.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
import copy
import gc
import random
from argparse import ArgumentParser
from pathlib import Path
import numpy as np
import torch
from omegaconf import OmegaConf
from torch.cuda.amp import GradScaler
from torchvision.transforms import ToPILImage
from tqdm import tqdm
from models.warp_inpaint_model import WarpInpaintModel
from util.finetune_utils import finetune_depth_model, finetune_decoder
from util.general_utils import apply_depth_colormap, save_video
def evaluate(model):
fps = model.config["save_fps"]
save_root = Path(model.run_dir)
save_dict = {
"images": torch.cat(model.images, dim=0),
"images_orig_decoder": torch.cat(model.images_orig_decoder, dim=0),
"masks": torch.cat(model.masks, dim=0),
"disparities": torch.cat(model.disparities, dim=0),
"depths": torch.cat(model.depths, dim=0),
"cameras": model.cameras_extrinsics,
}
torch.save(save_dict, save_root / "results.pt")
if not model.config["use_splatting"]:
model.save_mesh("full_mesh")
video = (255 * torch.cat(model.images, dim=0)).to(torch.uint8).detach().cpu()
video_reverse = (255 * torch.cat(model.images[::-1], dim=0)).to(torch.uint8).detach().cpu()
save_video(video, save_root / "output.mp4", fps=fps)
save_video(video_reverse, save_root / "output_reverse.mp4", fps=fps)
def evaluate_epoch(model, epoch):
disparity = model.disparities[epoch]
disparity_colored = apply_depth_colormap(disparity[0].permute(1, 2, 0))
disparity_colored = disparity_colored.clone().permute(2, 0, 1).unsqueeze(0).float()
save_root = Path(model.run_dir) / "images"
save_root.mkdir(exist_ok=True, parents=True)
(save_root / "frames").mkdir(exist_ok=True, parents=True)
(save_root / "images_orig_decoder").mkdir(exist_ok=True, parents=True)
(save_root / "masks").mkdir(exist_ok=True, parents=True)
(save_root / "warped_images").mkdir(exist_ok=True, parents=True)
(save_root / "disparities").mkdir(exist_ok=True, parents=True)
ToPILImage()(model.images[epoch][0]).save(save_root / "frames" / f"{epoch}.png")
ToPILImage()(model.images_orig_decoder[epoch][0]).save(save_root / "images_orig_decoder" / f"{epoch}.png")
ToPILImage()(model.masks[epoch][0]).save(save_root / "masks" / f"{epoch}.png")
ToPILImage()(model.warped_images[epoch][0]).save(save_root / "warped_images" / f"{epoch}.png")
ToPILImage()(disparity_colored[0]).save(save_root / "disparities" / f"{epoch}.png")
if epoch == 0:
with open(Path(model.run_dir) / "config.yaml", "w") as f:
OmegaConf.save(model.config, f)
def run(config):
seed = config["seed"]
if seed == -1:
seed = np.random.randint(2 ** 32)
random.seed(seed)
np.random.seed(seed)
torch.manual_seed(seed)
print(f"running with seed: {seed}.")
model = WarpInpaintModel(config).to(config["device"])
evaluate_epoch(model, 0)
scaler = GradScaler(enabled=config["enable_mix_precision"])
for epoch in tqdm(range(1, config["frames"] + 1)):
if config["use_splatting"]:
warp_output = model.warp_splatting(epoch)
else:
warp_output = model.warp_mesh(epoch)
inpaint_output = model.inpaint(warp_output["warped_image"], warp_output["inpaint_mask"])
if config["finetune_decoder"]:
finetune_decoder(config, model, warp_output, inpaint_output)
model.update_images_masks(inpaint_output["latent"], warp_output["inpaint_mask"])
if config["finetune_depth_model"]:
# reload depth model
del model.depth_model
gc.collect()
torch.cuda.empty_cache()
model.depth_model = torch.hub.load("intel-isl/MiDaS", "DPT_Large").to(model.device)
finetune_depth_model(config, model, warp_output, epoch, scaler)
model.update_depth(model.images[epoch])
if not config["use_splatting"]:
# update mesh with the correct mask
if config["mask_opening_kernel_size"] > 0:
mesh_mask = 1 - torch.maximum(model.masks[epoch], model.masks_diffs[epoch - 1])
else:
mesh_mask = 1 - model.masks[epoch]
extrinsic = model.get_extrinsics(model.current_camera)
model.update_mesh(model.images[epoch], model.depths[epoch], mesh_mask > 0.5, extrinsic, epoch)
# reload decoder
model.vae.decoder = copy.deepcopy(model.decoder_copy)
model.images_orig_decoder.append(model.decode_latents(inpaint_output["latent"]).detach())
evaluate_epoch(model, epoch)
torch.cuda.empty_cache()
gc.collect()
evaluate(model)
if __name__ == "__main__":
parser = ArgumentParser()
parser.add_argument(
"--base-config",
default="./config/base-config.yaml",
help="Config path",
)
parser.add_argument(
"--example_config",
default="./config/example_configs/dungeon.yaml",
help="Config path",
)
args = parser.parse_args()
base_config = OmegaConf.load(args.base_config)
example_config = OmegaConf.load(args.example_config)
config = OmegaConf.merge(base_config, example_config)
run(config)