forked from AdrianAlan/PF-Jet-SSD
-
Notifications
You must be signed in to change notification settings - Fork 0
/
jet-ssd-eval.py
331 lines (269 loc) · 12.3 KB
/
jet-ssd-eval.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
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
import argparse
import torch
import torch.nn as nn
import torch.backends.cudnn as cudnn
import yaml
import warnings
from ssd.net import build_ssd
from tqdm import tqdm
from utils import *
warnings.filterwarnings(
action='ignore',
category=UserWarning,
module=r'.*'
)
def execute(model,
dataset,
im_size,
obj_size,
conf_threshold=10**-6,
batch_size=50,
max_distance=.1,
num_classes=3,
epsilon=10**-6,
text='Evaluating Network',
verbose=False):
results = [torch.empty(0, 6).cpu() for _ in range(num_classes)]
deltas = torch.empty((0, 5))
if verbose:
progress_bar = tqdm(total=len(dataset), desc=text)
for X, y, _, scalers in dataset:
y_pred = model(X).data
for i in range(batch_size):
all_detections = torch.empty((0, 6))
detections, targets = y_pred[i], y[i]
scaler = scalers[i]
for cid, dts in enumerate(detections):
# Filter detections above given threshold
dts = dts[dts[:, 0] > conf_threshold]
if dts.size(0) == 0:
continue
dx = (dts[:, 1] + dts[:, 3]).unsqueeze(1)/2
dy = (dts[:, 2] + dts[:, 4]).unsqueeze(1)/2
scores = dts[:, 0].unsqueeze(1)
regres = dts[:, 5].unsqueeze(1) * scaler
labels = cid*torch.ones(len(scores)).unsqueeze(1)
truths = torch.zeros(len(scores)).unsqueeze(1)
# Format: [x, y, label, score, truth, pt]
dts = torch.cat((dx, dy, labels, scores, truths, regres), 1)
all_detections = torch.cat((all_detections, dts))
# Sort by confidence
all_detections = all_detections[(-all_detections[:, 3]).argsort()]
for t in targets:
detected = False
tx = (t[0]+t[2])/2
ty = (t[1]+t[3])/2
tp = t[5] * scaler
for x, d in enumerate(all_detections):
delta_eta = (tx-d[0])
delta_phi = torch.min((ty-d[1]) % 1, (d[1]-ty) % 1)
distance = torch.sqrt(delta_eta**2+delta_phi**2)
if distance > max_distance:
continue
if d[2] == t[4]:
detected = True
# Angular resolution and regression data
deta = np.radians(1)*im_size[0]*delta_eta
dphi = np.radians(1)*im_size[1]*delta_phi
dpt = d[5] / (tp + epsilon)
dts = torch.Tensor([t[4], tp, deta, dphi, dpt])
deltas = torch.cat((deltas, dts.unsqueeze(0)))
all_detections[x][0] = tx
all_detections[x][1] = ty
all_detections[x][4] = 1
all_detections[x][5] = tp
break
if not detected:
fn = torch.Tensor([tx, ty, t[4], 0, 1, tp])
fn = fn.unsqueeze(0)
all_detections = torch.cat((all_detections, fn))
for c in range(num_classes):
dets = all_detections[all_detections[:, 2] == (c + 1)].cpu()
results[c] = torch.cat((results[c], dets)).cpu()
if args.verbose:
progress_bar.update(1)
if args.verbose:
progress_bar.close()
return results, deltas.cpu()
def execute_baseline(dataset,
im_size,
obj_size,
batch_size=50,
max_distance=.1,
num_classes=3,
text='Evaluating Baseline',
epsilon=10**-6,
verbose=False):
results = [torch.empty(0, 6).cpu() for _ in range(num_classes)]
deltas = torch.empty((0, 5))
if verbose:
progress_bar = tqdm(total=len(dataset), desc=text)
for _, y, baseline, scalers in dataset:
for i in range(batch_size):
targets, scaler = y[i], scalers[i]
all_baselines = baseline[i]
# Sort by confidence
all_baselines = all_baselines[(-all_baselines[:, 3]).argsort()]
for t in targets:
detected = False
tx = (t[0]+t[2])/2
ty = (t[1]+t[3])/2
tp = t[5] * scaler
for x, b in enumerate(all_baselines):
delta_eta = (tx-b[0])
delta_phi = torch.min((ty-b[1]) % 1, (b[1]-ty) % 1)
distance = torch.sqrt(delta_eta**2+delta_phi**2)
if distance > max_distance:
continue
if b[2] == t[4]:
detected = True
# Angular resolution and regression data
deta = np.radians(1)*im_size[0]*delta_eta
dphi = np.radians(1)*im_size[1]*delta_phi
dpt = b[5] / (tp + epsilon)
dts = torch.Tensor([t[4], tp, deta, dphi, dpt])
deltas = torch.cat((deltas, dts.unsqueeze(0)))
all_baselines[x][0] = tx
all_baselines[x][1] = ty
all_baselines[x][4] = 1
all_baselines[x][5] = tp
break
if not detected:
fn = torch.Tensor([tx, ty, t[4], 0, 1, tp])
fn = fn.unsqueeze(0)
all_baselines = torch.cat((all_baselines, fn))
for c in range(num_classes):
dets = all_baselines[all_baselines[:, 2] == (c + 1)].cpu()
results[c] = torch.cat((results[c], dets))
if args.verbose:
progress_bar.update(1)
if args.verbose:
progress_bar.close()
return results, deltas.cpu()
if __name__ == '__main__':
parser = argparse.ArgumentParser('Evaluate Jet Detection Model')
parser.add_argument('fpn', type=str,
help='Full Precision Network model name')
parser.add_argument('twn', type=str,
help='Ternary Weight Network model name')
parser.add_argument('int8', type=str,
help='int8 Network model name')
parser.add_argument('-c', '--config', action=IsValidFile, type=str,
help='Path to config file', default='ssd-config.yml')
parser.add_argument('-s', '--structure', action=IsValidFile, type=str,
help='Path to config file', default='net-config.yml')
parser.add_argument('-v', '--verbose', action='store_true',
help='Output verbosity')
args = parser.parse_args()
config = yaml.safe_load(open(args.config))
net_config = yaml.safe_load(open(args.structure))
dataset = config['dataset']['test'][0]
evaluation_pref = config['evaluation_pref']
ssd_settings = config['ssd_settings']
net_channels = net_config['network_channels']
batch_size = evaluation_pref['batch_size']
jet_names = evaluation_pref['names_classes']
num_workers = evaluation_pref['workers']
conf_threshold = ssd_settings['confidence_threshold']
input_dimensions = ssd_settings['input_dimensions']
jet_size = ssd_settings['object_size']
num_classes = ssd_settings['n_classes']
ssd_settings['n_classes'] += 1
max_distance = ssd_settings['max_distance']
plotting_results = []
plotting_deltas = []
logger = set_logging('Test_SSD',
'{}/PF-Jet-SSD-Test.log'.format(
config['output']['model']),
args.verbose)
logger.info('Testing baseline')
torch.set_default_tensor_type('torch.cuda.FloatTensor')
loader = get_data_loader(dataset,
batch_size,
num_workers,
input_dimensions,
jet_size,
cpu=False,
return_baseline=True,
return_pt=True,
shuffle=False)
base_results, base_deltas = execute_baseline(loader,
input_dimensions[1:],
jet_size,
batch_size,
max_distance,
num_classes,
text='Evaluating Baseline',
verbose=args.verbose)
for i, name in enumerate([args.fpn, args.twn, args.int8]):
logger.info('Testing {0}'.format(name))
path = '{}/{}.pth'.format(config['output']['model'], name)
if i == 2:
torch.set_default_tensor_type('torch.FloatTensor')
net = build_ssd(torch.device('cpu'),
ssd_settings,
net_channels,
inference=True,
int8=True)
net.qconfig = torch.quantization.get_default_qat_qconfig('fbgemm')
torch.quantization.prepare_qat(net, inplace=True)
net.load_weights(path)
net = net.cpu()
torch.quantization.convert(net.eval(), inplace=True)
else:
torch.set_default_tensor_type('torch.cuda.FloatTensor')
net = build_ssd(0, ssd_settings, net_channels, inference=True)
net.load_weights(path)
cudnn.benchmark = True
net = net.cuda()
net.eval()
if i == 0:
dummy_input = torch.unsqueeze(torch.randn(input_dimensions), 0)
mac = GetResources(net, dummy_input).profile() / 1e9
logger.info('Total OPS: {0:.3f}G'.format(mac))
logger.info('Total network parameters: {0}'.format(
sum(p.numel() for p in net.parameters() if p.requires_grad)))
loader = get_data_loader(dataset,
batch_size,
num_workers,
input_dimensions,
jet_size,
cpu=(i == 2),
return_baseline=True,
return_pt=True,
shuffle=False)
with torch.no_grad():
results, deltas = execute(net,
loader,
input_dimensions[1:],
jet_size,
conf_threshold=conf_threshold,
batch_size=batch_size,
max_distance=max_distance,
num_classes=num_classes,
text='Evaluating {}'.format(name),
verbose=args.verbose)
plotting_results.append(results)
plotting_deltas.append(deltas)
plot = Plotting(save_dir=config['output']['plots'])
ap, pr3, pr5 = plot.draw_precision_recall(base_results,
plotting_results[0],
plotting_results[1],
plotting_results[2],
jet_names)
models = ['Baseline', 'FPN', 'TWN', 'INT8']
for jap, jpr3, jpr5, name in zip(ap, pr3, pr5, models):
for x, jet in enumerate(jet_names):
print('Results for {0}, {1} jets.'.format(name, jet))
print('AP:{0:.3f}; P@R=.3:{1:.3f}; P@R=.5:{2:.3f}.'.format(
jap[x], jpr3[x], jpr5[x]))
plot.draw_precision_details(base_deltas,
plotting_results[0],
plotting_results[1],
plotting_results[2],
jet_names)
plot.draw_loc_delta(base_deltas,
plotting_deltas[0],
plotting_deltas[1],
plotting_deltas[2],
jet_names)