-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathdata.py
159 lines (135 loc) · 5.82 KB
/
data.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
from torch.utils.data import Dataset, DataLoader
import torch
import numpy as np
import pandas as pd
import h5py
import json
import psana
class PSANADataset(Dataset):
def __init__(self, df_path, subset="train", n=-1, shuffle=False):
self.df = pd.read_csv(df_path).query("subset == '{}'".format(subset))
if n > 0:
n = min(n, len(self.df))
self.df = self.df.sample(n=n)
if shuffle:
self.df = self.df.sample(frac=1.0)
self.n = len(self.df)
def __len__(self):
return self.n
def __getitem__(self, idx):
file_path, exp, run = self.df.iloc[idx][["path", "exp", "run"]]
return file_path, exp, run
class PSANAImage(Dataset):
def __init__(self, cxi_path, exp, run, normalize=True, downsample=1, debug=True,
max_cutoff=1024, mode="peaknet2020", shuffle=False, n=-1):
self.downsample = downsample
self.cxi = CXILabel(cxi_path)
self.detector = self.cxi.detector # "CxiDs2.0:Cspad.0"#
if n == -1:
self.n = len(self.cxi)
else:
self.n = min(n, len(self.cxi))
self.normalize = normalize
self.max_cutoff = max_cutoff
self.debug = debug
self.psana = PSANAReader(exp, run, self.detector)
self.psana.build()
self.mode = mode
if shuffle:
self.rand_idxs = np.random.permutation(len(self.cxi))
self.rand_idxs = self.rand_idxs[:self.n]
else:
self.rand_idxs = np.arange(self.n)
def __len__(self):
return self.n
def make_label(self, s, r, c, n_panels=32, h=24, w=49):
label = torch.zeros(n_panels, 3, h, w)
for i in range(n_panels):
my_r = r[s == i]
my_c = c[s == i]
for j in range(len(my_r)):
u = int(np.floor(my_r[j] / float(self.downsample)))
v = int(np.floor(my_c[j] / float(self.downsample)))
label[i, 0, u, v] = 1
label[i, 1, u, v] = np.fmod(my_r[j] / float(self.downsample), 1.0)
label[i, 2, u, v] = np.fmod(my_c[j] / float(self.downsample), 1.0)
return label
def make_yolo_labels(self, s, r, c, h_obj=7, w_obj=7):
n = r.shape[0]
cls = np.zeros((n,))
ww = w_obj * np.ones((n,))
hh = h_obj * np.ones((n,))
return [cls, s, r, c, hh, ww]
def close(self):
self.cxi.close()
self.psana.ds = None
def __getitem__(self, idx):
event, s, r, c = self.cxi[self.rand_idxs[idx]]
# print(event, "s", s, "r", r, "c", c)
img = self.psana.load_img(event)
if self.normalize:
img[img < 0] = 0
img = img / max(0.01, np.std(img))
img = img - np.mean(img)
# img = img / max(np.max(img), self.max_cutoff)
h_ds = int(np.ceil(img.shape[1] / float(self.downsample)))
w_ds = int(np.ceil(img.shape[2] / float(self.downsample)))
h_pad = int(h_ds * self.downsample)
w_pad = int(w_ds * self.downsample)
if self.mode == "peaknet2020":
img_tensor = torch.zeros(img.shape[0], h_pad, w_pad)
img_tensor[:, 0:img.shape[1], 0:img.shape[2]] = torch.from_numpy(img)
label_tensor = self.make_label(s, r, c, n_panels=img.shape[0], h=h_ds, w=w_ds)
return img_tensor, label_tensor
else: # YOLO mode
labels = self.make_yolo_labels(s, r, c)
return img, labels
class PSANAReader(object):
def __init__(self, exp, run, det_name="DsdCsPad"):
self.exp = exp
self.run = run
self.det_name = det_name
self.ds = None
self.det = None
self.this_run = None
self.times = None
def build(self):
self.ds = psana.DataSource("exp={}:run={}:idx".format(self.exp, self.run))
self.det = psana.Detector(self.det_name)
self.this_run = self.ds.runs().next()
self.times = self.this_run.times()
def load_img(self, event_idx):
evt = self.this_run.event(self.times[event_idx])
calib = self.det.calib(evt) * self.det.mask(evt, calib=True, status=True, edges=True,
central=True, unbond=True, unbondnbrs=True)
return calib
class CXILabel(Dataset):
def __init__(self, cxi_path, fmod=True):
self.f = h5py.File(cxi_path, "r")
self.nPeaks = self.f["entry_1/result_1/nPeaks"]
self.n_hits = len(self.nPeaks)
self.eventIdx = self.f["LCLS/eventNumber"][:self.n_hits]
self.peak_x_label = self.f['entry_1/result_1/peakXPosRaw'][:self.n_hits, :]
self.peak_y_label = self.f['entry_1/result_1/peakYPosRaw'][:self.n_hits, :]
self.peak_x_center = self.f['entry_1/result_1/peak2'][:self.n_hits, :]
self.peak_y_center = self.f['entry_1/result_1/peak1'][:self.n_hits, :]
self.peak_w = self.f['entry_1/result_1/peak4'][:self.n_hits, :]
self.peak_h = self.f['entry_1/result_1/peak3'][:self.n_hits, :]
self.detector = str(self.f["entry_1/instrument_1/detector_1/description"].value)
def __len__(self):
return self.n_hits
def __getitem__(self, idx):
my_npeaks = self.nPeaks[idx]
my_event_idx = self.eventIdx[idx]
# psana style
my_s = np.floor_divide(self.peak_y_label[idx, 0:my_npeaks], 185) \
+ 8 * np.floor_divide(self.peak_x_label[idx, 0:my_npeaks], 388)
my_r = np.fmod(self.peak_y_center[idx, 0:my_npeaks], 185.0)
my_c = np.fmod(self.peak_x_center[idx, 0:my_npeaks], 388.0)
return my_event_idx, my_s, my_r, my_c
# psocake style
# my_r = self.peak_y_center[idx,0:my_npeaks]
# my_c = self.peak_x_center[idx,0:my_npeaks]
# return (my_event_idx, my_r, my_c)
def close(self):
self.f.close()