/
eval_ablation_studies.py
executable file
·396 lines (338 loc) · 17.7 KB
/
eval_ablation_studies.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
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
#!/usr/bin/env python
# coding: utf-8
# Copyright (c) Nanjing University, Vision Lab.
# Last update:
# 2019.10.27
# 2019.11.14
# 2020.11.26
import os
import time
import numpy as np
import tensorflow as tf
import matplotlib.pylab as plt
import pandas as pd
import subprocess
import glob
import configparser
import argparse
import importlib
# from numba import cuda
tf.enable_eager_execution()
from process import preprocess, postprocess
# import models.model_voxception as model
from transform import compress_factorized, decompress_factorized
from transform import compress_hyper, decompress_hyper
from dataprocess.inout_bitstream import write_binary_files_factorized, read_binary_files_factorized
from dataprocess.inout_bitstream import write_binary_files_hyper, read_binary_files_hyper
os.environ['CUDA_VISIBLE_DEVICES']="0"
# set gpu.
cfg = tf.ConfigProto()
cfg.gpu_options.per_process_gpu_memory_fraction = 1.0
cfg.gpu_options.allow_growth = True
cfg.log_device_placement=True
# config.device_count={'gpu':0}
sess = tf.Session(config=cfg)
from myutils.pc_error_wrapper import pc_error
from myutils.pc_error_wrapper import get_points_number
def set_default_config(input_file, cfg_rootdir, resolution, mode, cube_size, modelname="models.model_voxception"):
filename = os.path.split(input_file)[-1][:-4]
config_file = os.path.join(cfg_rootdir, filename+'.ini')
config = configparser.ConfigParser()
if os.path.exists(config_file):
config.read(config_file)
print('config already exists.')
else:
config["DEFAULT"] = {"cube_size": cube_size, "min_num": 64, "resolution":resolution}
if mode=="factorized":
if modelname=="models.model_voxception":
config["R1"] = {"scale": 0.625, "ckpt_dir": './checkpoints/factorized/a2b3/'}
config["R2"] = {"scale": 1.0, "ckpt_dir": './checkpoints/factorized/a2b3/'}
config["R3"] = {"scale": 1.0, "ckpt_dir": './checkpoints/factorized/a4b3/'}
config["R4"] = {"scale": 1.0, "ckpt_dir": './checkpoints/factorized/a6b3/'}
config["R5"] = {"scale": 1.0, "ckpt_dir": './checkpoints/factorized/a10b3/'}
config["R6"] = {"scale": 1.0, "ckpt_dir": './checkpoints/factorized/a16b3/'}
if modelname=="models.model_simple":
config["R1"] = {"scale": 1.0, "ckpt_dir": './checkpoints/factorized/simple/a1b3/'}
config["R2"] = {"scale": 1.0, "ckpt_dir": './checkpoints/factorized/simple/a2b3/'}
config["R3"] = {"scale": 1.0, "ckpt_dir": './checkpoints/factorized/simple/a3b3/'}
config["R4"] = {"scale": 1.0, "ckpt_dir": './checkpoints/factorized/simple/a4b3/'}
config["R5"] = {"scale": 1.0, "ckpt_dir": './checkpoints/factorized/simple/a5b3/'}
config["R6"] = {"scale": 1.0, "ckpt_dir": './checkpoints/factorized/simple/a6b3/'}
elif mode=="hyper":
config["R1"] = {"scale": 5/8., "ckpt_dir": './checkpoints/hyper/a0.75b3/'}
config["R2"] = {"scale": 1.0, "ckpt_dir": './checkpoints/hyper/a0.75b3/'}
config["R3"] = {"scale": 1.0, "ckpt_dir": './checkpoints/hyper/a2b3/'}
config["R4"] = {"scale": 1.0, "ckpt_dir": './checkpoints/hyper/a3.5b3/'}
config["R5"] = {"scale": 1.0, "ckpt_dir": './checkpoints/hyper/a6b3/'}
config["R6"] = {"scale": 1.0, "ckpt_dir": './checkpoints/hyper/a10b3/'}
config["R7"] = {"scale": 1.0, "ckpt_dir": './checkpoints/hyper/a16b3/'}
config.write(open(config_file, 'w'))
print('initialize config.')
return config, config_file
def test_factorized(input_file, model, ckpt_dir, scale, cube_size, min_num, postfix=''):
# Pre-process
cubes, cube_positions, points_numbers = preprocess(input_file, scale, cube_size, min_num)
### Encoding
strings, min_v, max_v, shape = compress_factorized(cubes, model, ckpt_dir)
# Write files
filename = os.path.split(input_file)[-1][:-4]
print(filename)
rootdir = './compressed'+ postfix +'/'
bytes_strings, bytes_pointnums, bytes_cubepos = write_binary_files_factorized(
filename, strings.numpy(), points_numbers, cube_positions,
min_v.numpy(), max_v.numpy(), shape.numpy(), rootdir)
# Read files
strings_d, points_numbers_d, cube_positions_d, min_v_d, max_v_d, shape_d = \
read_binary_files_factorized(filename, rootdir)
# Decoding
cubes_d = decompress_factorized(strings_d, min_v_d, max_v_d, shape_d, model, ckpt_dir)
# bpp
N = get_points_number(input_file)
bpp = round(8*(bytes_strings + bytes_pointnums + bytes_cubepos)/float(N), 4)
bpp_strings = round(8*bytes_strings/float(N), 4)
bpp_pointsnums = round(8*bytes_pointnums/float(N) ,4)
bpp_cubepos = round(8*bytes_cubepos/float(N), 4)
bpp_strings_hyper = 0
bpp_strings_head = 0
bpps = [bpp, bpp_strings, bpp_strings_hyper, bpp_strings_head, bpp_pointsnums, bpp_cubepos]
return cubes_d, cube_positions_d, points_numbers_d, N, bpps
def test_hyper(input_file, model, ckpt_dir, scale, cube_size, min_num, postfix=''):
# Pre-process
cubes, cube_positions, points_numbers = preprocess(input_file, scale, cube_size, min_num)
### Encoding
y_strings, y_min_vs, y_max_vs, y_shape, z_strings, z_min_v, z_max_v, z_shape, x_ds = compress_hyper(cubes, model, ckpt_dir, True)
# Write files
filename = os.path.split(input_file)[-1][:-4]
print(filename)
rootdir = './compressed'+ postfix +'/'
bytes_strings, bytes_strings_head, bytes_strings_hyper, bytes_pointnums, bytes_cubepos = write_binary_files_hyper(
filename, y_strings.numpy(), z_strings.numpy(), points_numbers, cube_positions,
y_min_vs.numpy(), y_max_vs.numpy(), y_shape.numpy(),
z_min_v.numpy(), z_max_v.numpy(), z_shape.numpy(), rootdir)
# Read files
y_strings_d, z_strings_d, points_numbers_d, cube_positions_d, y_min_vs_d, y_max_vs_d, y_shape_d, z_min_v_d, z_max_v_d, z_shape_d = \
read_binary_files_hyper(filename, rootdir)
# Decoding
cubes_d = decompress_hyper(y_strings_d, y_min_vs_d.astype('int32'), y_max_vs_d.astype('int32'),
y_shape_d, z_strings_d, z_min_v_d, z_max_v_d, z_shape_d, model, ckpt_dir)
# cheat!!!
##############
cubes_d = x_ds
##############
# bpp
N = get_points_number(input_file)
bpp = round(8*(bytes_strings + bytes_strings_head + bytes_strings_hyper +
bytes_pointnums + bytes_cubepos)/float(N), 4)
bpp_strings = round(8*bytes_strings/float(N), 4)
bpp_strings_hyper = round(8*bytes_strings_hyper/float(N), 4)
bpp_strings_head = round(8*bytes_strings_head/float(N), 4)
bpp_pointsnums = round(8*bytes_pointnums/float(N) ,4)
bpp_cubepos = round(8*bytes_cubepos/float(N), 4)
bpps = [bpp, bpp_strings, bpp_strings_hyper, bpp_strings_head, bpp_pointsnums, bpp_cubepos]
return cubes_d, cube_positions_d, points_numbers_d, N, bpps
def select_optimal_rho(item, rhos,
input_file, output_file, input_file_n,
cubes_d, points_numbers_d, cube_positions_d,
scale, cube_size, res):
for i, rho in enumerate(rhos):
print('===== select rho =====')
postprocess(output_file, cubes_d, points_numbers_d, cube_positions_d, scale, cube_size, rho)
results = pc_error(input_file, output_file, input_file_n, res, show=False)
PSNR = float(results[item])
print('===== results: ', i, rho, item, PSNR)
if i == 0:
MAX_PSNR = 0
optimal_rho = rho
else:
MAX_PSNR = max(PSNR, MAX_PSNR)
if PSNR < MAX_PSNR:
break
else:
optimal_rho = rho
return optimal_rho
def cfg_post_process(config, config_file, rate,
input_file, output_file, input_file_n,
cubes_d, points_numbers_d, cube_positions_d,
scale, cube_size, res):
# select rho for optimal d1/d2 metrics.
if config.has_option(rate, 'rho_d1'):
rho_d1 = float(config.get(rate, 'rho_d1'))
else:
item = "mseF,PSNR (p2point)"
rhos = [0.8, 0.9, 1.0, 1.02, 1.05, 1.10, 1.15, 1.2, 1.25, 1.30, 1.40, 1.50, 1.75, 2.0, 2.5, 3.0]
rho_d1 = select_optimal_rho(item, rhos,
input_file, output_file, input_file_n,
cubes_d, points_numbers_d, cube_positions_d,
scale, cube_size, res)
config.set(rate, 'rho_d1', str(rho_d1))
config.write(open(config_file, 'w'))
if config.has_option(rate, 'rho_d2'):
rho_d2 = float(config.get(rate, 'rho_d2'))
else:
item = "mseF,PSNR (p2plane)"
rhos = [1.0, 0.98, 0.95, 0.92, 0.90, 0.88, 0.85, 0.82, 0.80, 0.75, 0.70, 0.65, 0.50, 0.40, 0.30]
rho_d2 = select_optimal_rho(item, rhos,
input_file, output_file, input_file_n,
cubes_d, points_numbers_d, cube_positions_d,
scale, cube_size, res)
config.set(rate, 'rho_d2', str(rho_d2))
config.write(open(config_file, 'w'))
return rho_d1, rho_d2
def collect_results(results, results_d1, results_d2, bpps, N, scale, rho_d1, rho_d2):
# bpp
results["ori_points"] = N
results["scale"] = scale
# results["cube_size"] = cube_size
# results["res"] = res
results["bpp"] = bpps[0]
results["bpp_strings"] = bpps[1]
results["bpp_strings_hyper"] = bpps[2]
results["bpp_strings_head"] = bpps[3]
results["bpp_pointsnums"] = bpps[4]
results["bpp_cubepos"] = bpps[5]
results["rho_d1"] = rho_d1
results["optimal D1 PSNR"] = results_d1["mseF,PSNR (p2point)"]
results["rho_d2"] = rho_d2
results["optimal D2 PSNR"] = results_d2["mseF,PSNR (p2plane)"]
print(results)
return results
def plot_results(all_results, filename, root_dir):
fig, ax = plt.subplots(figsize=(7.3, 4.2))
plt.plot(np.array(all_results["bpp"][:]), np.array(all_results["mseF,PSNR (p2point)"][:]),
label="D1", marker='x', color='red')
plt.plot(np.array(all_results["bpp"][:]), np.array(all_results["mseF,PSNR (p2plane)"][:]),
label="D2", marker='x', color = 'blue')
plt.plot(np.array(all_results["bpp"][:]), np.array(all_results["optimal D1 PSNR"][:]),
label="D1 (optimal)", marker='h', color='red', linestyle='-.')
plt.plot(np.array(all_results["bpp"][:]), np.array(all_results["optimal D2 PSNR"][:]),
label="D2 (optimal)", marker='h', color='blue', linestyle='-.')
plt.title(filename)
plt.xlabel('bpp')
plt.ylabel('PSNR')
plt.grid(ls='-.')
plt.legend(loc='lower right')
fig.savefig(os.path.join(root_dir, filename+'.png'))
return
def eval(input_file, rootdir, resolution, mode, cube_size, modelname, fixed_thres, postfix):
# model = 'model_voxception'
model = importlib.import_module(modelname)
filename = os.path.split(input_file)[-1][:-4]
output_file = filename + '_rec_' + postfix + '.ply'
input_file_n = input_file
csv_rootdir = os.path.join(rootdir, "csv")
if not os.path.exists(csv_rootdir):
os.makedirs(csv_rootdir)
cfg_rootdir = os.path.join(rootdir, "cfg")
if not os.path.exists(cfg_rootdir):
os.makedirs(cfg_rootdir)
# default config
config, config_file = set_default_config(input_file, cfg_rootdir, resolution, mode, cube_size, modelname)
cube_size = config.getint('DEFAULT', 'cube_size')
min_num = config.getint('DEFAULT', 'min_num')
res = config.getint('DEFAULT', 'resolution')
print('cube size:', cube_size, 'min num:', min_num, 'res:', res)
for index, rate in enumerate(config.sections()):
scale = float(config.get(rate, 'scale'))
ckpt_dir = str(config.get(rate, 'ckpt_dir'))
print('====================', 'config:', rate, 'scale:', scale, 'ckpt_dir:', ckpt_dir)
if mode=="factorized":
cubes_d, cube_positions, points_numbers, N, bpps = test_factorized(input_file, model, ckpt_dir, scale, cube_size, min_num, postfix)
elif mode == "hyper":
cubes_d, cube_positions, points_numbers, N, bpps = test_hyper(input_file, model, ckpt_dir, scale, cube_size, min_num, postfix)
cubes_d = cubes_d.numpy()
print("bpp:",bpps[0])
# select rho for optimal d1/d2 metrics.
if fixed_thres==None:
rho_d1, rho_d2 = cfg_post_process(config, config_file, rate,
input_file, output_file, input_file_n,
cubes_d, points_numbers, cube_positions,
scale, cube_size, res)
else:
rho_d1, rho_d2 = 1.0, 1.0
# metrics.
rho = 1.0
postprocess(output_file, cubes_d, points_numbers, cube_positions, scale, cube_size, rho, fixed_thres)
results = pc_error(input_file, output_file, input_file_n, res, show=False)
rho = rho_d1
postprocess(output_file, cubes_d, points_numbers, cube_positions, scale, cube_size, rho, fixed_thres)
results_d1 = pc_error(input_file, output_file, input_file_n, res, show=False)
rho = rho_d2
postprocess(output_file, cubes_d, points_numbers, cube_positions, scale, cube_size, rho, fixed_thres)
results_d2 = pc_error(input_file, output_file, input_file_n, res, show=False)
results = collect_results(results, results_d1, results_d2, bpps, N, scale, rho_d1, rho_d2)
if index == 0:
all_results = results.copy(deep=True)
else:
all_results = all_results.append(results, ignore_index=True)
# write to csv
print(all_results)
if not os.path.exists(csv_rootdir):
os.makedirs(csv_rootdir)
csv_name = os.path.join(csv_rootdir, filename + '.csv')
all_results.to_csv(csv_name, index=False)
# plot
plot_results(all_results, filename, csv_rootdir)
return all_results
def parse_args():
parser = argparse.ArgumentParser(
formatter_class=argparse.ArgumentDefaultsHelpFormatter)
parser.add_argument("--input", type=str, nargs='+', default='', dest="input")
parser.add_argument("--rootdir", type=str, default='./results/', dest="rootdir")
parser.add_argument("--resolution", type=int, default=1024, dest="resolution")
parser.add_argument("--mode", type=str, default='hyper', dest="mode")
# for ablation studies
parser.add_argument("--cube_size", type=int, default=64, dest="cube_size")
parser.add_argument("--modelname", default="models.model_voxception", help="(model_simple, model_voxception)", dest="modelname")
parser.add_argument("--fixed_thres", type=float, default=None, help="fixed threshold ", dest="fixed_thres")
parser.add_argument("--postfix", default="", help="", dest="postfix")
args = parser.parse_args()
print(args)
return args
if __name__ == "__main__":
args = parse_args()
if not os.path.exists(args.rootdir):
os.makedirs(args.rootdir)
# shapenet_filedirs = glob.glob("testdata/ShapeNet/*.ply")
# args.input = shapenet_filedirs
# print(shapenet_filedirs)
for input_file in sorted(args.input):
print(input_file)
all_results = eval(input_file, args.rootdir, args.resolution, args.mode, args.cube_size, args.modelname, args.fixed_thres, args.postfix)
"""
python eval_ablation_studies.py --input 'testdata/8iVFB/longdress_vox10_1300.ply' \
'testdata/8iVFB/loot_vox10_1200.ply' \
'testdata/8iVFB/redandblack_vox10_1550.ply' \
'testdata/8iVFB/soldier_vox10_0690.ply' \
'testdata/8iVFB/queen_0200.ply' \
--rootdir='results/hyper/' \
--mode="hyper" \
--postfix='hyper'
ablation studies:
1. cube_size:
python eval_ablation_studies.py --input 'testdata/8iVFB/longdress_vox10_1300.ply'
--rootdir='results/hyper/cube32'
--mode="hyper"
--postfix='hyper32'
--cube_size=32
python eval_ablation_studies.py --input 'testdata/8iVFB/longdress_vox10_1300.ply'
--rootdir='results/hyper/cube128'
--mode="hyper"
--postfix='hyper128'
--cube_size=128
2. compression netowork:
python eval_ablation_studies.py --input 'testdata/8iVFB/longdress_vox10_1300.ply' \
--rootdir='results/factorized/' \
--mode="factorized" \
--postfix='factorized'
python eval_ablation_studies.py --input 'testdata/8iVFB/longdress_vox10_1300.ply' \
--rootdir='results/factorized/simple_model/' \
--mode="factorized" \
--postfix='factorized_simple_model' \
--model=models.model_simple
3. fixed threshold:
python eveval_ablation_studiesal.py --input 'testdata/8iVFB/longdress_vox10_1300.ply' \
--rootdir='results/hyper/fixed_thres/' \
--mode="hyper" \
--postfix='hyper_fixed' \
--fixed_thres=0
"""