-
Notifications
You must be signed in to change notification settings - Fork 7
/
sample_batch.py
177 lines (130 loc) · 5.97 KB
/
sample_batch.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
import os
import os.path as osp
import json
import numpy as np
import cv2
import torch
from matplotlib import colors
import matplotlib.pyplot as plt
from moviepy.editor import ImageSequenceClip
COLOR_MAP = [[0, 0, 0],
[0.30196078431372547, 0.30196078431372547, 0.30196078431372547],
[0.1568627450980392, 0.7450980392156863, 0.1568627450980392],
[0.09803921568627451, 0.5882352941176471, 0.09803921568627451],
[0.0392156862745098, 0.4117647058823529, 0.0392156862745098],
[0.0392156862745098, 0.29411764705882354, 0.0392156862745098],
[0.9607843137254902, 0.9607843137254902, 0.0],
[0.9294117647058824, 0.6745098039215687, 0.0],
[0.9411764705882353, 0.43137254901960786, 0.0],
[0.6274509803921569, 0.0, 0.0],
[0.9058823529411765, 0.0, 1.0]]
PIXEL_SCALE = 255.0
BOUNDS = [0.0, 16.0, 31.0, 59.0, 74.0, 100.0, 133.0, 160.0, 181.0, 219.0, PIXEL_SCALE]
PIXEL_SCALE = 255.0
img_channel = 1
img_size = 128
in_T = 5
out_T = 20
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
def gray2color(image):
cmap = colors.ListedColormap(COLOR_MAP )
bounds = BOUNDS
norm = colors.BoundaryNorm(bounds, cmap.N)
colored_image = cmap(norm(image))
return colored_image
def vis_res(pred_seq, gt_seq, save_path, save_grays=False, save_colored=False):
# pred_seq: ndarray, [T, C, H, W], value range: [0, 1] float
if isinstance(pred_seq, torch.Tensor):
pred_seq = pred_seq.detach().cpu().numpy()
gt_seq = gt_seq.detach().cpu().numpy()
pred_seq = pred_seq.squeeze()
gt_seq = gt_seq.squeeze()
os.makedirs(save_path, exist_ok=True)
if save_grays:
os.makedirs(osp.join(save_path, 'pred'), exist_ok=True)
os.makedirs(osp.join(save_path, 'targets'), exist_ok=True)
for i, (pred, gt) in enumerate(zip(pred_seq, gt_seq)):
plt.imsave(osp.join(save_path, 'pred', f'{i}.png'), pred, cmap='gray', vmax=1.0, vmin=0.0)
plt.imsave(osp.join(save_path, 'targets', f'{i}.png'), gt, cmap='gray', vmax=1.0, vmin=0.0)
pred_seq = pred_seq * PIXEL_SCALE
pred_seq = pred_seq.astype(np.int16)
gt_seq = gt_seq * PIXEL_SCALE
gt_seq = gt_seq.astype(np.int16)
colored_pred = np.array([gray2color(pred_seq[i]) for i in range(len(pred_seq))], dtype=np.float64)
colored_gt = np.array([gray2color(gt_seq[i]) for i in range(len(gt_seq))],dtype=np.float64)
if save_colored:
os.makedirs(osp.join(save_path, 'pred_colored'), exist_ok=True)
os.makedirs(osp.join(save_path, 'targets_colored'), exist_ok=True)
for i, (pred, gt) in enumerate(zip(colored_pred, colored_gt)):
plt.imsave(osp.join(save_path, 'pred_colored', f'{i}.png'), pred)
plt.imsave(osp.join(save_path, 'targets_colored', f'{i}.png'), gt)
clip = ImageSequenceClip(list(colored_pred * 255), fps=4)
clip.write_gif(osp.join(save_path, 'pred.gif'), fps=4, verbose=False)
clip = ImageSequenceClip(list(colored_gt * 255), fps=4)
clip.write_gif(osp.join(save_path, 'targets.gif'), fps=4, verbose=False)
def build_model(chkpt_path):
from models.phydnet import get_model
kwargs = {
"in_shape": (img_channel, img_size, img_size),
"T_in": in_T,
"T_out": out_T,
"device": device
}
backbone = get_model(**kwargs)
from diffcast import get_model
kwargs = {
'img_channels' : img_channel,
'dim' : 64,
'dim_mults' : (1,2,4,8),
'T_in': in_T,
'T_out': out_T,
'sampling_timesteps': 250,
}
diff_model = get_model(**kwargs)
diff_model.load_backbone(backbone)
diff_model = diff_model.to(device)
data = torch.load(chkpt_path, map_location=device)
print(data.keys())
diff_model.load_state_dict(data['model'])
print(f"Loaded model from {chkpt_path}")
return diff_model
def read_data(data_path):
def read_seqs(sample_path):
targets = []
for path in sorted(os.listdir(os.path.join(sample_path, "targets")), key=lambda x: int(x.split('.')[0])):
# print(path)
if path.endswith(".png"):
targets.append(cv2.imread(os.path.join(sample_path, 'targets',path), cv2.IMREAD_GRAYSCALE))
targets = np.array(targets)
inputs = []
for path in sorted(os.listdir(os.path.join(sample_path, "preds")), key=lambda x: int(x.split('.')[0])):
# print(path)
if path.endswith(".png"):
inputs.append(cv2.imread(os.path.join(sample_path, 'preds', path), cv2.IMREAD_GRAYSCALE))
inputs = np.array(inputs)
return targets, inputs
dirs = sorted(os.listdir(data_path))
targets, inputs = [], []
for dir in dirs:
seq_path = os.path.join(data_path, dir)
target_seq, input_seq = read_seqs(seq_path)
targets.append(target_seq)
inputs.append(input_seq)
inputs = np.array(inputs)
inputs = torch.from_numpy(inputs).unsqueeze(2).float() / PIXEL_SCALE
targets = np.array(targets)
targets = torch.from_numpy(targets).unsqueeze(2).float() / PIXEL_SCALE
print("Data shape: ", inputs.shape, targets.shape)
print("Data range: ", inputs.min(), inputs.max(), targets.min(), targets.max())
return inputs.to(device), targets.to(device)
cur_dir = os.path.dirname(os.path.abspath(__file__))
chkpt_path = os.path.join(cur_dir, "resources", "diffcast_phydnet_sevir128.pt")
data_path = os.path.join(cur_dir,"resources", "seqs")
model = build_model(chkpt_path)
input_batch, target_batch = read_data(data_path)
pred_batch, mus, rs = model.sample(input_batch, T_out=out_T)
pred_batch = pred_batch.clamp(0, 1.0)
for b in range(len(pred_batch)):
save_name = osp.join(cur_dir, "resources", "preds",f"demo_{b}")
vis_res(pred_batch[b], target_batch[b], save_path=save_name, save_grays=True, save_colored=True)
vis_res(mus[b], target_batch[b], save_path=save_name+"_mu", save_colored=True)