-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathtrain.py
executable file
·315 lines (271 loc) · 15.8 KB
/
train.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
#
# Copyright (C) 2023, Inria
# GRAPHDECO research group, https://team.inria.fr/graphdeco
# All rights reserved.
#
# This software is free for non-commercial, research and evaluation use
# under the terms of the LICENSE.md file.
#
# For inquiries contact [email protected]
#
import os
import torch
from random import randint
from utils.loss_utils import l1_loss, l1_loss_map, Scale_balance_loss, scale_regulation_loss, scale_region_regulation_loss, get_trained_seg
from gaussian_renderer import render, network_gui
import sys
from scene import Scene, GaussianModel
from utils.general_utils import safe_state
import uuid
from tqdm import tqdm
from utils.image_utils import psnr
from argparse import ArgumentParser, Namespace
from arguments import ModelParams, PipelineParams, OptimizationParams
try:
from torch.utils.tensorboard import SummaryWriter
TENSORBOARD_FOUND = True
except ImportError:
TENSORBOARD_FOUND = False
import torch.nn.functional as F
from models.networks import CNN_decoder, CNN_scale_decoder
from scene.dataset_readers import read_sam_clip_feature
from segment_anything import sam_model_registry
from preprocess import OpenCLIPNetworkConfig, OpenCLIPNetwork
def create_scale_map(single_scale, feature_map_shape):
scale_values = {
"s": [1, 0, 0],
"m": [0, 1, 0],
"l": [0, 0, 1],
"mix": [1.0 / 3.0, 1.0 / 3.0, 1.0 / 3.0]
}
assert single_scale in scale_values, "Invalid scale value"
scale_map = torch.tensor(scale_values[single_scale], dtype=torch.float32, device='cuda')
return scale_map.unsqueeze(-1).unsqueeze(-1).repeat(1, feature_map_shape[1], feature_map_shape[2])
def training(dataset, opt, pipe, testing_iterations, saving_iterations, checkpoint_iterations, checkpoint, debug_from, scale_balance_iteration, scale_regulation_iteration, render_novel_view_iteration, novel_view_interval, feature_mode, single_scale):
device0='cuda'
device1='cpu'
first_iter = 0
tb_writer = prepare_output_and_logger(dataset)
gaussians = GaussianModel(dataset.sh_degree)
scene = Scene(dataset, gaussians, shuffle=False)
# 2D semantic feature map CNN decoder
viewpoint_stack = scene.getTrainCameras()
camnum_orig=len(viewpoint_stack)
viewpoint_cam0 = viewpoint_stack[0]
feature_out_dim = viewpoint_cam0.img_embed.shape[1]
render_h,render_w = viewpoint_cam0.image_height, viewpoint_cam0.image_width
print("render img with H,W:",render_h,",",render_w)
# feature decoding
feature_in_dim = int(feature_out_dim / 32)
if dataset.speedup:
cnn_decoder = CNN_decoder(feature_in_dim, feature_out_dim)
cnn_decoder = cnn_decoder.to(device0)
cnn_decoder_optimizer = torch.optim.Adam(cnn_decoder.parameters(), lr=0.0001)
# scale decoding
cnn_scale_decoder = CNN_scale_decoder(feature_in_dim, 3)
cnn_scale_decoder = cnn_scale_decoder.to(device0)
cnn_scale_decoder_optimizer = torch.optim.Adam(cnn_scale_decoder.parameters(), lr=0.0001)
original_h, original_w = viewpoint_cam0.semantic_feature_height, viewpoint_cam0.semantic_feature_width
gaussians.training_setup(opt)
if checkpoint: # continue from checkpoint
(model_params, first_iter) = torch.load(checkpoint)
if len(model_params) == 12 and feature_mode:
first_iter = 0
else: # feature field
# load feature decoder ckpt
cnn_decoder_ckpt=torch.load(os.path.join(dataset.model_path, "decoder_chkpnt" + str(first_iter) + ".pth"))
cnn_decoder.load_state_dict(cnn_decoder_ckpt['module_state_dict'])
cnn_decoder_optimizer.load_state_dict(cnn_decoder_ckpt['optimizer_state_dict'])
# load cscale decoder ckpt
cnn_scale_decoder_ckpt=torch.load(os.path.join(dataset.model_path, "scale_decoder_chkpnt" + str(first_iter) + ".pth"))
cnn_scale_decoder.load_state_dict(cnn_scale_decoder_ckpt['module_state_dict'])
cnn_scale_decoder_optimizer.load_state_dict(cnn_scale_decoder_ckpt['optimizer_state_dict'])
gaussians.restore(model_params, opt)
print("number of gaussians",gaussians._xyz.shape)
# set other parameters
bg_color = [1, 1, 1] if dataset.white_background else [0, 0, 0]
background = torch.tensor(bg_color, dtype=torch.float32, device="cuda")
iter_start = torch.cuda.Event(enable_timing = True)
iter_end = torch.cuda.Event(enable_timing = True)
viewpoint_stack = None
ema_loss_for_log = 0.0
progress_bar = tqdm(range(first_iter, opt.iterations), desc="Training progress")
first_iter += 1
for iteration in range(first_iter, opt.iterations + 1):
if network_gui.conn == None:
network_gui.try_connect()
while network_gui.conn != None:
try:
net_image_bytes = None
custom_cam, do_training, pipe.convert_SHs_python, pipe.compute_cov3D_python, keep_alive, scaling_modifer = network_gui.receive()
if custom_cam != None:
net_image = render(custom_cam, gaussians, pipe, background, scaling_modifer)["render"]
net_image_bytes = memoryview((torch.clamp(net_image, min=0, max=1.0) * 255).byte().permute(1, 2, 0).contiguous().cpu().numpy()) #将net_image处理并通过memoryview提供一个直接访问这些数据的接口 clamp()将net_image的值截断至[0,1]
network_gui.send(net_image_bytes, dataset.source_path)
if do_training and ((iteration < int(opt.iterations)) or not keep_alive):
break
except Exception as e:
network_gui.conn = None
iter_start.record()
gaussians.update_learning_rate(iteration)
# Every 1000 its we increase the levels of SH up to a maximum degree
if iteration % 1000 == 0:
gaussians.oneupSHdegree()
# Pick a random Camera
if not viewpoint_stack:
viewpoint_stack = scene.getTrainCameras().copy()
select_idx = randint(0, len(viewpoint_stack)-1)
viewpoint_cam = viewpoint_stack.pop(select_idx)
# Render
if (iteration - 1) == debug_from:
pipe.debug = True
render_pkg = render(viewpoint_cam, gaussians, pipe, background, feature_mode=feature_mode)
feature_map, viewspace_point_tensor, visibility_filter, radii = render_pkg["render"], render_pkg["viewspace_points"], render_pkg["visibility_filter"], render_pkg["radii"]
# 固定scale
if single_scale:
scale_map = create_scale_map(single_scale, feature_map.shape)
else:
scale_map=cnn_scale_decoder(feature_map.detach()) # 3,H,W
# Loss
seg_map_trained = get_trained_seg(viewpoint_cam.seg_map.to(device0), scale_map)
feature_reionvar_loss = scale_region_regulation_loss(feature_map, seg_map_trained, mix_seg=True) # feature 区域分布数据 方差
# scale_reionvar_loss = scale_region_regulation_loss(scale_map, seg_map_trained, mix_seg=True) # scale 区域分布数据 方差
scale_CE_loss = scale_regulation_loss(scale_map)
if dataset.speedup:
feature_map = cnn_decoder(feature_map)
if iteration < scale_balance_iteration: # L_distill
gt_feature_map, seg_mask = read_sam_clip_feature(viewpoint_cam.img_embed.to(device0), viewpoint_cam.seg_map.to(device0), scale_map)
Ll1_feature = l1_loss(feature_map * seg_mask, gt_feature_map * seg_mask)
else: # L_r-distill
gt_feature_map, seg_mask = read_sam_clip_feature(viewpoint_cam.img_embed.to(device0), viewpoint_cam.seg_map.to(device0), scale_map)
Ll1_feature = l1_loss_map(feature_map * seg_mask, gt_feature_map * seg_mask)
Ll1_feature = Scale_balance_loss(Ll1_feature, seg_map_trained.to(device0), seg_mask.squeeze(0).to(device0), mix_seg=True)
if iteration < scale_regulation_iteration:
loss = 1.0 * Ll1_feature + 0.001 * scale_CE_loss
else:
loss = 1.0 * Ll1_feature + 0.002 * scale_CE_loss + 0.1 * feature_reionvar_loss
loss.backward()
iter_end.record()
with torch.no_grad():
# Progress bar
ema_loss_for_log = 0.4 * loss.item() + 0.6 * ema_loss_for_log
if iteration % 10 == 0:
progress_bar.set_postfix({"Loss": f"{ema_loss_for_log:.{7}f}"})
progress_bar.update(10)
if iteration == opt.iterations:
progress_bar.close()
# Log and save
training_report(tb_writer, iteration, Ll1_feature, feature_reionvar_loss, torch.mean(scale_map[0]),torch.mean(scale_map[1]),torch.mean(scale_map[2]), loss, l1_loss,
iter_start.elapsed_time(iter_end), testing_iterations, scene, render, (pipe, background))
if (iteration in saving_iterations):
print("\n[ITER {}] Saving Gaussians".format(iteration))
mem = torch.cuda.max_memory_allocated() / 1024**3
print(f"Max memory used: {mem:.2f} GB")
scene.save(iteration)
# Densification
if iteration < opt.densify_until_iter and not feature_mode:
# Keep track of max radii in image-space for pruning
gaussians.max_radii2D[visibility_filter] = torch.max(gaussians.max_radii2D[visibility_filter], radii[visibility_filter])
# gaussians.add_densification_stats(viewspace_point_tensor, visibility_filter)
gaussians.add_densification_stats(viewspace_point_tensor, visibility_filter, feature_map.shape[2], feature_map.shape[1])
if iteration > opt.densify_from_iter and iteration % opt.densification_interval == 0:
size_threshold = 20 if iteration > opt.opacity_reset_interval else None
gaussians.densify_and_prune(opt.densify_grad_threshold, 0.005, scene.cameras_extent, size_threshold) # 增删高斯
if iteration % opt.opacity_reset_interval == 0 or (dataset.white_background and iteration == opt.densify_from_iter):
gaussians.reset_opacity()
# Optimizer step
if iteration < opt.iterations:
gaussians.optimizer.step()
gaussians.optimizer.zero_grad(set_to_none = True)
if dataset.speedup:
cnn_decoder_optimizer.step()
cnn_decoder_optimizer.zero_grad(set_to_none = True)
cnn_scale_decoder_optimizer.step()
cnn_scale_decoder_optimizer.zero_grad(set_to_none = True)
if (iteration in checkpoint_iterations):
print("\n[ITER {}] Saving Checkpoint".format(iteration))
torch.save((gaussians.capture(), iteration), scene.model_path + "/chkpnt" + str(iteration) + ".pth")
print("\n[ITER {}] Saving feature decoder ckpt".format(iteration))
if dataset.speedup:
torch.save({'module_state_dict':cnn_decoder.state_dict(),
'optimizer_state_dict':cnn_decoder_optimizer.state_dict()
},
scene.model_path + "/decoder_chkpnt" + str(iteration) + ".pth")
torch.save({'module_state_dict':cnn_scale_decoder.state_dict(),
'optimizer_state_dict':cnn_scale_decoder_optimizer.state_dict()
},
scene.model_path + "/scale_decoder_chkpnt" + str(iteration) + ".pth")
def prepare_output_and_logger(args):
if not args.model_path:
if os.getenv('OAR_JOB_ID'):
unique_str=os.getenv('OAR_JOB_ID')
else:
unique_str = str(uuid.uuid4())
args.model_path = os.path.join("./output/", unique_str[0:10])
# Set up output folder
print("Output folder: {}".format(args.model_path))
os.makedirs(args.model_path, exist_ok = True)
with open(os.path.join(args.model_path, "cfg_args"), 'w') as cfg_log_f:
cfg_log_f.write(str(Namespace(**vars(args))))
# Create Tensorboard writer
tb_writer = None
if TENSORBOARD_FOUND:
tb_writer = SummaryWriter(args.model_path)
else:
print("Tensorboard not available: not logging progress")
return tb_writer
def training_report(tb_writer, iteration, Ll1_feature, feature_reionvar_loss, scale_s, sclae_m, scale_l, loss, l1_loss, elapsed, testing_iterations, scene : Scene, renderFunc, renderArgs):
if tb_writer:
# tb_writer.add_scalar('train_loss_patches/l1_loss', Ll1.item(), iteration)
tb_writer.add_scalar('train_loss_patches/l1_loss_feature', Ll1_feature.item(), iteration)
tb_writer.add_scalar('train_loss_patches/total_loss', loss.item(), iteration)
tb_writer.add_scalar('train_loss_patches/feature_reionvar_loss', feature_reionvar_loss.item(), iteration)
tb_writer.add_scalar('iter_time', elapsed, iteration)
tb_writer.add_scalar('scale_patchs/subpart',scale_s.item(), iteration)
tb_writer.add_scalar('scale_patchs/part',sclae_m.item(), iteration)
tb_writer.add_scalar('scale_patchs/whole',scale_l.item(), iteration)
if __name__ == "__main__":
# Set up command line argument parser
parser = ArgumentParser(description="Training script parameters")
lp = ModelParams(parser)
op = OptimizationParams(parser)
pp = PipelineParams(parser)
parser.add_argument('--ip', type=str, default="127.0.0.1")
parser.add_argument('--port', type=int, default=6009)
parser.add_argument('--debug_from', type=int, default=-1)
parser.add_argument('--detect_anomaly', action='store_true', default=False)
parser.add_argument("--test_iterations", nargs="+", type=int, default=[15_000, 30_000])
parser.add_argument("--save_iterations", nargs="+", type=int, default=[15_000, 30_000])
parser.add_argument("--quiet", action="store_true")
parser.add_argument("--checkpoint_iterations", nargs="+", type=int, default=[15_000, 30_000])
parser.add_argument("--start_checkpoint", type=str, default = None)
parser.add_argument('--scale_balance_iteration', type=int, default=1)
parser.add_argument('--scale_regulation_iteration', type=int, default=15001)
parser.add_argument('--render_novel_view_iteration',type=int, default=99999)
parser.add_argument('--novel_view_interval',type=int,default=150)
parser.add_argument('--feature_mode', action='store_true', help='use feature replace RGB')
parser.add_argument('--sam_ckpt_path', type=str, default="ckpts/sam_vit_h_4b8939.pth")
parser.add_argument("--novel_view", action='store_true')
parser.add_argument("--single_scale",type=str, choices=['s', 'm', 'l', 'mix'], default = None) # s | m | l
args = parser.parse_args(sys.argv[1:])
args.save_iterations.append(args.iterations)
print("Optimizing " + args.model_path)
# Initialize system state (RNG)
safe_state(args.quiet)
# Initialize SAM & CLIP model
if args.novel_view:
CLIP_model = OpenCLIPNetwork(OpenCLIPNetworkConfig)
sam = sam_model_registry["vit_h"](checkpoint=args.sam_ckpt_path).to('cuda')
else:
CLIP_model = None
sam = None
# empty cache
torch.cuda.empty_cache()
# Start GUI server, configure and run training
network_gui.init(args.ip, args.port)
torch.autograd.set_detect_anomaly(args.detect_anomaly)
training(lp.extract(args), op.extract(args), pp.extract(args),
args.test_iterations, args.save_iterations, args.checkpoint_iterations, args.start_checkpoint,
args.debug_from, args.scale_balance_iteration, args.scale_regulation_iteration,args.render_novel_view_iteration,args.novel_view_interval,args.feature_mode,args.single_scale)
# All done
print("\nTraining complete.")