forked from m-srk/SGAE
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy patheval_mem.py
198 lines (178 loc) · 9.2 KB
/
eval_mem.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
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from captioning.data.dataloaderraw import *
# from captioning.data.dataloader import *
from dataloader_extend import *
# from six.moves import cPickle
import _pickle as cPickle
import json
import numpy as np
import time
import os
from six.moves import cPickle
# import _pickle as cPickle
# import opts
import models
import eval_utils_mem
from captioning.data.dataloaderraw import *
import argparse
import misc.utils as utils
import torch
os.environ["CUDA_VISIBLE_DEVICES"] = "1"
device = torch.device('cpu')
print(device, "in", __file__)
# Input arguments and options
parser = argparse.ArgumentParser()
# Input paths
parser.add_argument('--model', type=str, default='',
help='path to model to evaluate')
parser.add_argument('--cnn_model', type=str, default='resnet101',
help='resnet101, resnet152')
parser.add_argument('--infos_path', type=str, default='',
help='path to infos to evaluate')
# Basic options
parser.add_argument('--batch_size', type=int, default=0,
help='if > 0 then overrule, otherwise load from checkpoint.')
parser.add_argument('--num_images', type=int, default=-1,
help='how many images to use when periodically evaluating the loss? (-1 = all)')
parser.add_argument('--language_eval', type=int, default=0,
help='Evaluate language as well (1 = yes, 0 = no)? BLEU/CIDEr/METEOR/ROUGE_L? requires coco-caption code from Github.')
parser.add_argument('--dump_images', type=int, default=1,
help='Dump images into vis/imgs folder for vis? (1=yes,0=no)')
parser.add_argument('--dump_json', type=int, default=1,
help='Dump json with predictions into vis folder? (1=yes,0=no)')
parser.add_argument('--dump_path', type=int, default=0,
help='Write image paths along with predictions into vis json? (1=yes,0=no)')
# Sampling options
parser.add_argument('--sample_max', type=int, default=1,
help='1 = sample argmax words. 0 = sample from distributions.')
parser.add_argument('--max_ppl', type=int, default=0,
help='beam search by max perplexity or max probability.')
parser.add_argument('--beam_size', type=int, default=2,
help='used when sample_max = 1, indicates number of beams in beam search. Usually 2 or 3 works well. More is not better. Set this to 1 for faster runtime but a bit worse performance.')
parser.add_argument('--group_size', type=int, default=1,
help='used for diverse beam search. if group_size is 1, then it\'s normal beam search')
parser.add_argument('--diversity_lambda', type=float, default=0.5,
help='used for diverse beam search. Usually from 0.2 to 0.8. Higher value of lambda produces a more diverse list')
parser.add_argument('--temperature', type=float, default=1.0,
help='temperature when sampling from distributions (i.e. when sample_max = 0). Lower = "safer" predictions.')
parser.add_argument('--decoding_constraint', type=int, default=0,
help='If 1, not allowing same word in a row')
# For evaluation on a folder of images:
parser.add_argument('--image_folder', type=str, default='',
help='If this is nonempty then will predict on the images in this folder path')
parser.add_argument('--image_root', type=str, default='',
help='In case the image paths have to be preprended with a root path to an image folder')
# For evaluation on MSCOCO images from some split:
parser.add_argument('--input_fc_dir', type=str, default='',
help='path to the h5file containing the preprocessed dataset')
parser.add_argument('--input_att_dir', type=str, default='',
help='path to the h5file containing the preprocessed dataset')
parser.add_argument('--input_box_dir', type=str, default='',
help='path to the h5file containing the preprocessed dataset')
parser.add_argument('--input_label_h5', type=str, default='',
help='path to the h5file containing the preprocessed dataset')
parser.add_argument('--input_json', type=str, default='',
help='path to the json file containing additional info and vocab. empty = fetch from model checkpoint.')
parser.add_argument('--split', type=str, default='test',
help='if running on MSCOCO images, which split to use: val|test|train')
parser.add_argument('--coco_json', type=str, default='',
help='if nonempty then use this file in DataLoaderRaw (see docs there). Used only in MSCOCO test evaluation, where we have a specific json file of only test set images.')
# misc
parser.add_argument('--id', type=str, default='',
help='an id identifying this run/job. used only if language_eval = 1 for appending to intermediate files')
parser.add_argument('--verbose_beam', type=int, default=1,
help='if we need to print out all beam search beams.')
parser.add_argument('--verbose_loss', type=int, default=0,
help='if we need to calculate loss.')
parser.add_argument('--verbose', type=int, default=1,
help='if we need to print out all beam search beams.')
parser.add_argument('--use_rela', type=int, default=0,
help='whether to use relationship matrix.')
parser.add_argument('--use_gru', type=int, default=0,
help='whether to use relationship matrix.')
parser.add_argument('--use_gfc', type=int, default=0,
help='whether to use relationship matrix.')
parser.add_argument('--use_ssg', type=int, default=0,
help='If use ssg')
parser.add_argument('--sg_dict_path', type=str, default='data/sg_dict_extend.npz',
help='path to the sentence scene graph directory')
parser.add_argument('--gru_t', type=int, default=4,
help='the numbers of gru will iterate')
parser.add_argument('--index_eval', type=int, default=1,
help='whether eval or not')
parser.add_argument('--input_rela_dir', type=str, default='data/cocobu_sg_img',
help='path to the directory containing the relationships of att feats')
parser.add_argument('--input_ssg_dir', type=str, default='data/coco_spice_sg',
help='path to the directory containing the ground truth sentence scene graph')
parser.add_argument('--output_cap_path', type=str, default='0',
help='file which save the result')
parser.add_argument('--training_mode', type=int, default=0,
help='training_mode')
parser.add_argument('--memory_cell_path', type=str, default='0',
help='memory_cell_path')
opt = parser.parse_args()
# Load infos
with open(opt.infos_path, 'rb') as f:
try:
infos = cPickle.load(f)
except Exception as e:
print('Exception caught ->\n', e)
# override and collect parameters
if len(opt.input_fc_dir) == 0:
opt.input_fc_dir = infos['opt'].input_fc_dir
opt.input_att_dir = infos['opt'].input_att_dir
opt.input_box_dir = infos['opt'].input_box_dir
opt.input_label_h5 = infos['opt'].input_label_h5
if len(opt.input_json) == 0:
opt.input_json = infos['opt'].input_json
if opt.batch_size == 0:
opt.batch_size = infos['opt'].batch_size
if len(opt.id) == 0:
opt.id = infos['opt'].id
opt.ssg_dict_path = 'data/ssg_dict.npz'
ignore = ["id", "batch_size", "beam_size", "start_from", "language_eval",
"use_rela", "memory_cell_path", "input_rela_dir", "--ssg_dict_path", 'model']
beam_size = opt.beam_size
for k in vars(infos['opt']).keys():
if k != 'model':
if k not in ignore:
if k in vars(opt):
assert vars(opt)[k] == vars(infos['opt'])[
k], k + ' option not consistent'
else:
# copy over options from model
vars(opt).update({k: vars(infos['opt'])[k]})
vocab = infos['vocab'] # ix -> word mapping
# Setup the model
model = models.setup(opt)
model.load_state_dict(torch.load(opt.model))
model.to(device)
model.eval()
crit = utils.LanguageModelCriterion()
# Create the Data Loader instance
if len(opt.image_folder) == 0:
loader = DataLoader(opt)
else:
loader = DataLoaderRaw({'folder_path': opt.image_folder,
'coco_json': opt.coco_json,
'batch_size': opt.batch_size,
'cnn_model': opt.cnn_model})
# When eval using provided pretrained model, the vocab may be different from what you have in your cocotalk.json
# So make sure to use the vocab in infos file.
loader.ix_to_word = infos['vocab']
# Set sample options
loss, split_predictions, lang_stats = eval_utils_mem.eval_split(model, crit, loader, opt.training_mode,
vars(opt))
print('loss: ', loss)
if lang_stats:
print(lang_stats)
text_file = open('res'+opt.id+'.txt', "aw")
text_file.write('{0}\n'.format(opt.model))
text_file.write('beam_size {0}\n'.format(beam_size))
text_file.write('{0}\n'.format(lang_stats))
text_file.close()
if opt.dump_json == 1:
# dump the json
json.dump(split_predictions, open('vis/vis.json', 'w'))