-
Notifications
You must be signed in to change notification settings - Fork 0
/
test.py
215 lines (189 loc) · 9.24 KB
/
test.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
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
import os
import jittor as jt
from PIL import Image
import numpy as np
from tqdm import tqdm
from jnerf.ops.code_ops import *
from jnerf.dataset.dataset import jt_srgb_to_linear, jt_linear_to_srgb
from jnerf.utils.config import get_cfg, save_cfg
from jnerf.utils.registry import build_from_cfg,NETWORKS,SCHEDULERS,DATASETS,OPTIMS,SAMPLERS,LOSSES
from jnerf.models.losses.mse_loss import img2mse, mse2psnr
from jnerf.dataset import camera_path
import cv2
from jnerf.utils.config import init_cfg
class Runner():
def __init__(self):
self.cfg = get_cfg()
if self.cfg.fp16 and jt.flags.cuda_archs[0] < 70:
print("Warning: Sm arch is lower than sm_70, fp16 is not supported. Automatically use fp32 instead.")
self.cfg.fp16 = False
if not os.path.exists(self.cfg.log_dir):
os.makedirs(self.cfg.log_dir)
self.exp_name = self.cfg.exp_name
self.dataset = {}
self.dataset["train"] = build_from_cfg(self.cfg.dataset.test, DATASETS)
# self.dataset["train"] = None
self.cfg.dataset_obj = self.dataset["train"]
if self.cfg.dataset.val:
self.dataset["val"] = build_from_cfg(self.cfg.dataset.test, DATASETS)
# self.dataset["val"] = None
else:
self.dataset["val"] = self.dataset["train"]
self.dataset["test"] = None
self.model = build_from_cfg(self.cfg.model, NETWORKS)
self.cfg.model_obj = self.model
self.sampler = build_from_cfg(self.cfg.sampler, SAMPLERS)
self.cfg.sampler_obj = self.sampler
self.optimizer = build_from_cfg(self.cfg.optim, OPTIMS, params=self.model.parameters())
self.optimizer = build_from_cfg(self.cfg.expdecay, OPTIMS, nested_optimizer=self.optimizer)
self.ema_optimizer = build_from_cfg(self.cfg.ema, OPTIMS, params=self.model.parameters())
self.loss_func = build_from_cfg(self.cfg.loss, LOSSES)
self.background_color = self.cfg.background_color
self.tot_train_steps = self.cfg.tot_train_steps
self.n_rays_per_batch = self.cfg.n_rays_per_batch
self.using_fp16 = self.cfg.fp16
self.save_path = os.path.join(self.cfg.log_dir, self.exp_name)
if not os.path.exists(self.save_path):
os.makedirs(self.save_path)
if self.cfg.ckpt_path and self.cfg.ckpt_path is not None:
self.ckpt_path = self.cfg.ckpt_path
else:
self.ckpt_path = os.path.join(self.save_path, "params.pkl")
if self.cfg.load_ckpt:
self.load_ckpt(self.ckpt_path)
else:
self.start=0
self.cfg.m_training_step = 0
self.val_freq = 4096
self.image_resolutions = self.dataset["train"].resolution
self.W = self.image_resolutions[0]
self.H = self.image_resolutions[1]
def test(self, load_ckpt=False):
if load_ckpt:
assert os.path.exists(self.ckpt_path), "ckpt file does not exist: "+self.ckpt_path
self.load_ckpt(self.ckpt_path)
if self.dataset["test"] is None:
self.dataset["test"] = build_from_cfg(self.cfg.dataset.test, DATASETS)
if not os.path.exists(os.path.join(self.save_path, "test")):
os.makedirs(os.path.join(self.save_path, "test"))
mse_list=self.render_test(save_path=os.path.join(self.save_path, "test"))
if self.dataset["test"].have_img:
tot_psnr=0
for mse in mse_list:
tot_psnr += mse2psnr(mse)
print("TOTAL TEST PSNR===={}".format(tot_psnr/len(mse_list)))
def load_ckpt(self, path):
print("Loading ckpt from:", path)
ckpt = jt.load(path)
self.start = ckpt['global_step']
self.model.load_state_dict(ckpt['model'])
if self.using_fp16:
self.model.set_fp16()
self.sampler.load_state_dict(ckpt['sampler'])
self.optimizer.load_state_dict(ckpt['optimizer'])
nested = ckpt['nested_optimizer']['defaults']['param_groups'][0]
for pg in self.optimizer._nested_optimizer.param_groups:
for i in range(len(pg["params"])):
pg["values"][i] = jt.array(nested["values"][i])
pg["m"][i] = jt.array(nested["m"][i])
ema = ckpt['ema_optimizer']['defaults']['param_groups'][0]
for pg in self.ema_optimizer.param_groups:
for i in range(len(pg["params"])):
pg["values"][i] = jt.array(ema["values"][i])
self.ema_optimizer.steps = ckpt['ema_optimizer']['defaults']['steps']
def render_test(self, save_img=True, save_path=None):
# save_path = "result"
if save_path is None:
save_path = self.save_path
mse_list = []
print("rendering testset...")
for img_i in tqdm(range(0, self.dataset["test"].n_images, 1)):
# for img_i in tqdm(range(0, 10, 1)):
with jt.no_grad():
imgs = []
for i in range(1):
simg, img_tar = self.render_img(dataset_mode="test", img_id=img_i)
imgs.append(simg)
img = np.stack(imgs, axis=0).mean(0)
if save_img:
self.save_img(save_path + f"/{self.exp_name}_r_{img_i}.png", img)
if self.dataset["test"].have_img:
self.save_img(save_path + f"/{self.exp_name}_gt_{img_i}.png", img_tar)
mse_list.append(img2mse(
jt.array(img),
jt.array(img_tar)).item())
return mse_list
def save_img(self, path, img):
if isinstance(img, np.ndarray):
ndarr = (img * 255 + 0.5).clip(0, 255).astype('uint8')
elif isinstance(img, jt.Var):
ndarr = (img * 255 + 0.5).clamp(0, 255).uint8().numpy()
im = Image.fromarray(ndarr)
im.save(path)
def render_img(self, dataset_mode="train", img_id=None):
W, H = self.image_resolutions
H = int(H)
W = int(W)
if img_id is None:
img_id = np.random.randint(0, self.dataset[dataset_mode].n_images, [1])[0]
img_ids = jt.zeros([H * W], 'int32') + img_id
else:
img_ids = jt.zeros([H * W], 'int32') + img_id
rays_o_total, rays_d_total, rays_pix_total = self.dataset[dataset_mode].generate_rays_total_test(
img_ids, W, H)
rays_pix_total = rays_pix_total.unsqueeze(-1)
pixel = 0
imgs = np.empty([H * W + self.n_rays_per_batch, 3])
for pixel in range(0, W * H, self.n_rays_per_batch):
end = pixel + self.n_rays_per_batch
rays_o = rays_o_total[pixel:end]
rays_d = rays_d_total[pixel:end]
if end > H * W:
rays_o = jt.concat(
[rays_o, jt.ones([end - H * W] + rays_o.shape[1:], rays_o.dtype)], dim=0)
rays_d = jt.concat(
[rays_d, jt.ones([end - H * W] + rays_d.shape[1:], rays_d.dtype)], dim=0)
pos, dir = self.sampler.sample(img_ids, rays_o, rays_d)
network_outputs = self.model(pos, dir)
rgb = self.sampler.rays2rgb(network_outputs, inference=True)
imgs[pixel:end] = rgb.numpy()
imgs = imgs[:H * W].reshape(H, W, 3)
imgs_tar = jt.array(self.dataset[dataset_mode].image_data[img_id]).reshape(H, W, 4)
imgs_tar = imgs_tar[..., :3] * imgs_tar[..., 3:] + jt.array(self.background_color) * (1 - imgs_tar[..., 3:])
imgs_tar = imgs_tar.detach().numpy()
jt.gc()
return imgs, imgs_tar
exp_name={'1':'_Car','2':'_Coffee','3':'_Easyship','4':'_Scar','5':'_Scarf'}
for i in range(0,5,1):
i = str(i+1)
expnumber = exp_name[i]
init_cfg("./projects/ngp/configs/ngp_comp" + expnumber + ".py")
test = Runner()
test.test(True)
for i in range(0,10):
number = str(i)
filePath = "logs/Easyship/test/"+"Easyship_r_"+number+".png"
img = cv2.imread(filePath, flags=1)
dst = cv2.fastNlMeansDenoisingColored(img, None, 60, 60, 7, 21)
saveFile = "result/" + "Easyship_r_" + number + ".png"
cv2.imwrite(saveFile, dst)
filePath = "logs/Car/test/" + "Car_r_" + number + ".png"
img = cv2.imread(filePath, flags=1)
dst = cv2.fastNlMeansDenoisingColored(img, None, 60, 60, 7, 21)
saveFile = "result/" + "Car_r_" + number + ".png"
cv2.imwrite(saveFile, dst)
filePath = "logs/Coffee/test/" + "Coffee_r_" + number + ".png"
img = cv2.imread(filePath, flags=1)
dst = cv2.fastNlMeansDenoisingColored(img, None, 3.7, 3.7, 7, 21)
saveFile = "result/" + "Coffee_r_" + number + ".png"
cv2.imwrite(saveFile, dst)
filePath = "logs/Scar/test/" + "Scar_r_" + number + ".png"
img = cv2.imread(filePath, flags=1)
dst = cv2.fastNlMeansDenoisingColored(img, None, 3.8, 3.8, 7, 21)
saveFile = "result/" + "Scar_r_" + number + ".png"
cv2.imwrite(saveFile, dst)
filePath = "logs/Scarf/test/" + "Scarf_r_" + number + ".png"
img = cv2.imread(filePath, flags=1)
dst = cv2.fastNlMeansDenoisingColored(img, None, 0.17, 0.17, 7, 21)
saveFile = "result/" + "Scarf_r_" + number + ".png"
cv2.imwrite(saveFile, dst)