-
Notifications
You must be signed in to change notification settings - Fork 2
/
waypoint_predictor.py
327 lines (271 loc) · 14.8 KB
/
waypoint_predictor.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
import torch
import argparse
from dataloader import RGBDepthPano
from image_encoders import RGBEncoder, DepthEncoder
from TRM_net import BinaryDistPredictor_TRM, TRM_predict
from eval import waypoint_eval
import os
import glob
import utils
import random
from utils import nms
from utils import print_progress
from tensorboardX import SummaryWriter
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
def setup(args):
torch.manual_seed(0)
random.seed(0)
exp_log_path = './checkpoints/%s/'%(args.EXP_ID)
os.makedirs(exp_log_path, exist_ok=True)
exp_log_path = './checkpoints/%s/snap/'%(args.EXP_ID)
os.makedirs(exp_log_path, exist_ok=True)
class Param():
def __init__(self):
self.parser = argparse.ArgumentParser(description='Train waypoint predictor')
self.parser.add_argument('--EXP_ID', type=str, default='test_0')
self.parser.add_argument('--TRAINEVAL', type=str, default='train', help='trian or eval mode')
self.parser.add_argument('--VIS', type=int, default=0, help='visualize predicted hearmaps')
# self.parser.add_argument('--LOAD_EPOCH', type=int, default=None, help='specific an epoch to load for eval')
self.parser.add_argument('--ANGLES', type=int, default=24)
self.parser.add_argument('--NUM_IMGS', type=int, default=24)
self.parser.add_argument('--NUM_CLASSES', type=int, default=12)
self.parser.add_argument('--MAX_NUM_CANDIDATES', type=int, default=5)
self.parser.add_argument('--PREDICTOR_NET', type=str, default='TRM', help='TRM only')
self.parser.add_argument('--EPOCH', type=int, default=10)
self.parser.add_argument('--BATCH_SIZE', type=int, default=2)
self.parser.add_argument('--LEARNING_RATE', type=float, default=1e-4)
self.parser.add_argument('--WEIGHT', type=int, default=0, help='weight the target map')
self.parser.add_argument('--TRM_LAYER', default=2, type=int, help='number of TRM hidden layers')
self.parser.add_argument('--TRM_NEIGHBOR', default=2, type=int, help='number of attention mask neighbor')
self.parser.add_argument('--HEATMAP_OFFSET', default=2, type=int, help='an offset determined by image FoV and number of images')
self.parser.add_argument('--HIDDEN_DIM', default=768, type=int)
self.args = self.parser.parse_args()
def predict_waypoints(args):
print('\nArguments', args)
log_dir = './checkpoints/%s/tensorboard/'%(args.EXP_ID)
writer = SummaryWriter(log_dir=log_dir)
''' networks '''
rgb_encoder = RGBEncoder(resnet_pretrain=True, trainable=False).to(device)
depth_encoder = DepthEncoder(resnet_pretrain=True, trainable=False).to(device)
if args.PREDICTOR_NET == 'TRM':
print('\nUsing TRM predictor')
print('HIDDEN_DIM default to 768')
args.HIDDEN_DIM = 768
predictor = BinaryDistPredictor_TRM(args=args,
hidden_dim=args.HIDDEN_DIM, n_classes=args.NUM_CLASSES).to(device)
''' load navigability (gt waypoints, obstacles and weights) '''
navigability_dict = utils.load_gt_navigability(
'./training_data/%s_*_mp3d_waypoint_twm0.2_obstacle_first_withpos.json'%(args.ANGLES))
''' dataloader for rgb and depth images '''
train_img_dir = './gen_training_data/rgbd_fov90/train/*/*.pkl'
traindataloader = RGBDepthPano(args, train_img_dir, navigability_dict)
eval_img_dir = './gen_training_data/rgbd_fov90/val_unseen/*/*.pkl'
evaldataloader = RGBDepthPano(args, eval_img_dir, navigability_dict)
if args.TRAINEVAL == 'train':
trainloader = torch.utils.data.DataLoader(traindataloader,
batch_size=args.BATCH_SIZE, shuffle=True, num_workers=4)
evalloader = torch.utils.data.DataLoader(evaldataloader,
batch_size=args.BATCH_SIZE, shuffle=False, num_workers=4)
''' optimization '''
criterion_bcel = torch.nn.BCEWithLogitsLoss(reduction='none')
criterion_mse = torch.nn.MSELoss(reduction='none')
params = list(predictor.parameters())
optimizer = torch.optim.AdamW(params, lr=args.LEARNING_RATE)
''' training loop '''
if args.TRAINEVAL == 'train':
print('\nTraining starts')
best_val_1 = {"avg_wayscore": 0.0, "log_string": '', "update":False}
best_val_2 = {"avg_pred_distance": 10.0, "log_string": '', "update":False}
for epoch in range(args.EPOCH): # loop over the dataset multiple times
sum_loss = 0.0
rgb_encoder.eval()
depth_encoder.eval()
predictor.train()
for i, data in enumerate(trainloader):
scan_ids = data['scan_id']
waypoint_ids = data['waypoint_id']
rgb_imgs = data['rgb'].to(device)
depth_imgs = data['depth'].to(device)
''' checking image orientation '''
# from PIL import Image
# from matplotlib import pyplot
# import numpy as np
# # import pdb; pdb.set_trace()
# out_img = np.swapaxes(
# np.swapaxes(
# data['no_trans_rgb'][0].cpu().numpy(), 1,2),
# 2, 3)
# for kk, out_img_i in enumerate(out_img):
# im = Image.fromarray(out_img_i)
# im.save("./play/%s.png"%(kk))
# pyplot.imsave("./play/mpl_%s.png"%(kk), out_img_i)
# out_depth = data['no_trans_depth'][0].cpu().numpy() * 255
# out_depth = out_depth.astype(np.uint8)
# for kk, out_depth_i in enumerate(out_depth):
# im = Image.fromarray(out_depth_i)
# im.save("./play/depth_%s.png"%(kk))
''' processing observations '''
rgb_feats = rgb_encoder(rgb_imgs) # (BATCH_SIZE*ANGLES, 2048)
depth_feats = depth_encoder(depth_imgs) # (BATCH_SIZE*ANGLES, 128, 4, 4)
''' learning objectives '''
target, obstacle, weight, _, _ = utils.get_gt_nav_map(
args.ANGLES, navigability_dict, scan_ids, waypoint_ids)
target = target.to(device)
obstacle = obstacle.to(device)
weight = weight.to(device)
if args.PREDICTOR_NET == 'TRM':
vis_logits = TRM_predict('train', args,
predictor, rgb_feats, depth_feats)
loss_vis = criterion_mse(vis_logits, target)
if args.WEIGHT:
loss_vis = loss_vis * weight
total_loss = loss_vis.sum() / vis_logits.size(0) / args.ANGLES
total_loss.backward()
optimizer.step()
sum_loss += total_loss.item()
print_progress(i+1, len(trainloader), prefix='Epoch: %d/%d'%((epoch+1),args.EPOCH))
writer.add_scalar("Train/Loss", sum_loss/(i+1), epoch)
print('Train Loss: %.5f' % (sum_loss/(i+1))) # (epoch+1),args.EPOCH
''' evaluation - inference '''
# print('Evaluation ...')
sum_loss = 0.0
predictions = {'sample_id': [],
'source_pos': [], 'target_pos': [],
'probs': [], 'logits': [],
'target': [], 'obstacle': [], 'sample_loss': []}
rgb_encoder.eval()
depth_encoder.eval()
predictor.eval()
for i, data in enumerate(evalloader):
scan_ids = data['scan_id']
waypoint_ids = data['waypoint_id']
sample_id = data['sample_id']
rgb_imgs = data['rgb'].to(device)
depth_imgs = data['depth'].to(device)
target, obstacle, weight, \
source_pos, target_pos = utils.get_gt_nav_map(
args.ANGLES, navigability_dict, scan_ids, waypoint_ids)
target = target.to(device)
obstacle = obstacle.to(device)
weight = weight.to(device)
''' processing observations '''
rgb_feats = rgb_encoder(rgb_imgs) # (BATCH_SIZE*ANGLES, 2048)
depth_feats = depth_encoder(depth_imgs) # (BATCH_SIZE*ANGLES, 128, 4, 4)
if args.PREDICTOR_NET == 'TRM':
vis_probs, vis_logits = TRM_predict('eval', args,
predictor, rgb_feats, depth_feats)
overall_probs = vis_probs
overall_logits = vis_logits
loss_vis = criterion_mse(vis_logits, target)
if args.WEIGHT:
loss_vis = loss_vis * weight
sample_loss = loss_vis.sum(-1).sum(-1) / args.ANGLES
total_loss = loss_vis.sum() / vis_logits.size(0) / args.ANGLES
sum_loss += total_loss.item()
predictions['sample_id'].append(sample_id)
predictions['source_pos'].append(source_pos)
predictions['target_pos'].append(target_pos)
predictions['probs'].append(overall_probs.tolist())
predictions['logits'].append((overall_logits.tolist()))
predictions['target'].append(target.tolist())
predictions['obstacle'].append(obstacle.tolist())
predictions['sample_loss'].append(target.tolist())
print('Eval Loss: %.5f' % (sum_loss/(i+1)))
results = waypoint_eval(args, predictions)
writer.add_scalar("Evaluation/Loss", sum_loss/(i+1), epoch)
writer.add_scalar("Evaluation/p_waypoint_openspace", results['p_waypoint_openspace'], epoch)
writer.add_scalar("Evaluation/p_waypoint_obstacle", results['p_waypoint_obstacle'], epoch)
writer.add_scalar("Evaluation/avg_wayscore", results['avg_wayscore'], epoch)
writer.add_scalar("Evaluation/avg_pred_distance", results['avg_pred_distance'], epoch)
log_string = 'Epoch %s '%(epoch)
for key, value in results.items():
if key != 'candidates':
log_string += '{} {:.5f} | '.format(str(key), value)
print(log_string)
# save checkpoint
if results['avg_wayscore'] > best_val_1['avg_wayscore']:
checkpoint_save_path = './checkpoints/%s/snap/check_val_best_avg_wayscore'%(args.EXP_ID) #, epoch+1
utils.save_checkpoint(epoch+1, predictor, optimizer, checkpoint_save_path)
print('New best avg_wayscore result found, checkpoint saved to %s'%(checkpoint_save_path))
best_val_1['avg_wayscore'] = results['avg_wayscore']
best_val_1['log_string'] = log_string
checkpoint_reg_save_path = './checkpoints/%s/snap/check_latest'%(args.EXP_ID) #, epoch+1
utils.save_checkpoint(epoch+1, predictor, optimizer, checkpoint_reg_save_path)
print('Best avg_wayscore result til now: ', best_val_1['log_string'])
if results['avg_pred_distance'] < best_val_2['avg_pred_distance']:
checkpoint_save_path = './checkpoints/%s/snap/check_val_best_avg_pred_distance'%(args.EXP_ID) #, epoch+1
utils.save_checkpoint(epoch+1, predictor, optimizer, checkpoint_save_path)
print('New best avg_pred_distance result found, checkpoint saved to %s'%(checkpoint_save_path))
best_val_2['avg_pred_distance'] = results['avg_pred_distance']
best_val_2['log_string'] = log_string
checkpoint_reg_save_path = './checkpoints/%s/snap/check_latest'%(args.EXP_ID) #, epoch+1
utils.save_checkpoint(epoch+1, predictor, optimizer, checkpoint_reg_save_path)
print('Best avg_pred_distance result til now: ', best_val_2['log_string'])
elif args.TRAINEVAL == 'eval':
''' evaluation - inference (with a bit mixture-of-experts) '''
print('\nEvaluation mode, please doublecheck EXP_ID and LOAD_EPOCH')
checkpoint_load_path = './checkpoints/%s/snap/check_val_best_avg_wayscore'%(args.EXP_ID) #args.LOAD_EPOCH
epoch, predictor, optimizer = utils.load_checkpoint(
predictor, optimizer, checkpoint_load_path)
sum_loss = 0.0
predictions = {'sample_id': [],
'source_pos': [], 'target_pos': [],
'probs': [], 'logits': [],
'target': [], 'obstacle': [], 'sample_loss': []}
rgb_encoder.eval()
depth_encoder.eval()
predictor.eval()
for i, data in enumerate(evalloader):
if args.VIS and i == 5:
break
scan_ids = data['scan_id']
waypoint_ids = data['waypoint_id']
sample_id = data['sample_id']
rgb_imgs = data['rgb'].to(device)
depth_imgs = data['depth'].to(device)
target, obstacle, weight, \
source_pos, target_pos = utils.get_gt_nav_map(
args.ANGLES, navigability_dict, scan_ids, waypoint_ids)
target = target.to(device)
obstacle = obstacle.to(device)
weight = weight.to(device)
''' processing observations '''
rgb_feats = rgb_encoder(rgb_imgs) # (BATCH_SIZE*ANGLES, 2048)
depth_feats = depth_encoder(depth_imgs) # (BATCH_SIZE*ANGLES, 128, 4, 4)
''' predicting the waypoint probabilities '''
if args.PREDICTOR_NET == 'TRM':
vis_probs, vis_logits = TRM_predict('eval', args,
predictor, rgb_feats, depth_feats)
overall_probs = vis_probs
overall_logits = vis_logits
loss_vis = criterion_mse(vis_logits, target)
if args.WEIGHT:
loss_vis = loss_vis * weight
sample_loss = loss_vis.sum(-1).sum(-1) / args.ANGLES
total_loss = loss_vis.sum() / vis_logits.size(0) / args.ANGLES
sum_loss += total_loss.item()
predictions['sample_id'].append(sample_id)
predictions['source_pos'].append(source_pos)
predictions['target_pos'].append(target_pos)
predictions['probs'].append(overall_probs.tolist())
predictions['logits'].append(overall_logits.tolist())
predictions['target'].append(target.tolist())
predictions['obstacle'].append(obstacle.tolist())
predictions['sample_loss'].append(target.tolist())
print('Eval Loss: %.5f' % (sum_loss/(i+1)))
results = waypoint_eval(args, predictions)
log_string = 'Epoch %s '%(epoch)
for key, value in results.items():
if key != 'candidates':
log_string += '{} {:.5f} | '.format(str(key), value)
print(log_string)
print('Evaluation Done')
else:
RunningModeError
if __name__ == "__main__":
param = Param()
args = param.args
setup(args)
if args.VIS:
assert args.TRAINEVAL == 'eval'
predict_waypoints(args)