-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathinference.py
227 lines (198 loc) · 11.2 KB
/
inference.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
import time
import os
from contextlib import contextmanager
import torch
from tqdm import tqdm
import numpy as np
import copy
import cv2
from rankevaluation.SASOR import evalu as rank_evalu
from rankevaluation.mae_fmeasure_2 import evalu as mf_evalu
import pickle as pkl
from detectron2.data import build_detection_test_loader
from mask2former.data.dataset_mappers.assr_dataset_mapper import AssrDatasetMapper
from mask2former.data.dataset_mappers.irsr_dataset_mapper import IrsrDatasetMapper
from mask2former.data.dataset_mappers.sifr_dataset_mapper import SIFRdataDatasetMapper
from detectron2.structures import Boxes, ImageList, Instances, BitMasks
from pycocotools import mask as coco_mask
from rankevaluation.sor_eval import doOffcialSorInference
import detectron2.utils.comm as comm
from mask2former.data.datasets.register_sifr import get_sifr_dicts
from mask2former.data.datasets.register_assr import get_assr_dicts
from mask2former.data.datasets.register_irsr import get_irsr_dicts
from detectron2.data import DatasetFromList, MapDataset
from torch.utils.data import DataLoader,SequentialSampler
import math
from thop import profile
def find_all_indexes(lst, value):
indexes = []
for i in range(len(lst)):
if lst[i] == value:
indexes.append(i)
return indexes
def convert_coco_poly_to_mask(segmentations, height, width):
masks = []
for polygons in segmentations:
rles = coco_mask.frPyObjects(polygons, height, width)
mask = coco_mask.decode(rles)
if len(mask.shape) < 3:
mask = mask[..., None]
mask = torch.as_tensor(mask, dtype=torch.uint8)
mask = mask.any(dim=2)
masks.append(mask)
if masks:
masks = torch.stack(masks, dim=0)
else:
masks = torch.zeros((0, height, width), dtype=torch.uint8)
return masks
@contextmanager
def inference_context(model):
"""
A context where the model is temporarily changed to eval mode,
and restored to previous mode afterwards.
Args:
model: a torch Module
"""
training_mode = model.training
model.eval()
yield
model.train(training_mode)
def trivial_batch_collator(batch):
"""
A batch collator that does nothing.
"""
return batch
def inference(cfg, model,model_name,model_root_dir,datasetmode):
if comm.is_main_process():
dataset=cfg.EVALUATION.DATASET
limited=cfg.EVALUATION.LIMITED
dataPath=cfg.EVALUATION.DATAPATH
if dataset=="assr":
SOR_DATASETPATH=dataPath+"ASSR/"
print('------Evaluation based on ASSR dataset!------')
assrdataset = get_assr_dicts(root=SOR_DATASETPATH, mode=datasetmode)
assrdatasetlist = DatasetFromList(assrdataset, copy=False)
assrdatasetlist = MapDataset(assrdatasetlist, AssrDatasetMapper(cfg, False))
dataloader = DataLoader(assrdatasetlist, batch_size=1, shuffle=False, num_workers=0,collate_fn=trivial_batch_collator)
elif dataset=='irsr':
SOR_DATASETPATH = dataPath+"IRSR/"
print('------Evaluation based on IRSR dataset!------')
irsrdataset = get_irsr_dicts(root=SOR_DATASETPATH, mode=datasetmode)
irsrdatasetlist = DatasetFromList(irsrdataset, copy=False)
irsrdatasetlist = MapDataset(irsrdatasetlist, IrsrDatasetMapper(cfg, False))
dataloader = DataLoader(irsrdatasetlist, batch_size=1, shuffle=False, num_workers=0,collate_fn=trivial_batch_collator)
elif dataset=='sifr':
SOR_DATASETPATH = dataPath+"SIFR/"
print('------Evaluation based on SIFR dataset!------')
sifrdataset = get_sifr_dicts(root=SOR_DATASETPATH, mode=datasetmode)
sifrdatasetlist = DatasetFromList(sifrdataset, copy=False)
sifrdatasetlist = MapDataset(sifrdatasetlist, SIFRdataDatasetMapper(cfg, False))
dataloader = DataLoader(sifrdatasetlist, batch_size=1, shuffle=False, num_workers=0,collate_fn=trivial_batch_collator)
ourputdir='./evaluationResult/'+model_name.split('.')[0]+'/'
saliencymapPath = os.path.join(ourputdir,f'GeneratedSaliencyMaps/', model_name+'-ResultThres-'+str(cfg.EVALUATION.RESULT_THRESHOLD)+f'Limited-{limited}'+'/')
print("------saliencyMapPath is :"+ saliencymapPath)
if not os.path.exists(saliencymapPath):
os.makedirs(saliencymapPath)
with inference_context(model), torch.no_grad():
res = []
for idx, inputs in enumerate(dataloader):
predictions=model(inputs)
img_height = inputs[0]["height"]
img_width = inputs[0]["width"]
if "instances" in predictions[-1]:
instances = predictions[-1]["instances"].to("cpu")
pred_instances = Instances(instances.image_size)
flag = False
for index in range(len(instances)):
score = instances[index].scores
if score > cfg.EVALUATION.RESULT_THRESHOLD:
if flag == False:
pred_instances = instances[index]
flag = True
else:
pred_instances = Instances.cat([pred_instances, instances[index]])
gt_masks_polygon = []
gt_ranks = []
for ins in inputs[0]['annotations']:
gt_ranks.append(ins['gt_rank'])
segm = []
for seg in ins['segmentation']:
segm.append(np.asarray(seg))
gt_masks_polygon.append(segm)
gt_masks = convert_coco_poly_to_mask(gt_masks_polygon, img_height, img_width).cpu().data.numpy()
name = inputs[0]["file_name"].split('/')[-1]
if flag:
pred_masks = pred_instances.pred_masks.cpu().data.numpy()
pred_ranks = pred_instances.pred_rank.cpu().data.numpy()
print('len of pred_rank')
if limited:
if dataset == 'assr':
if len(pred_ranks) > 5:
sorted_data = sorted(enumerate(pred_ranks), key=lambda x: x[1], reverse=True)
top5_indices = [index for index, value in sorted_data[:5]]
mask = [i in top5_indices for i in range(len(pred_ranks))]
pred_masks = pred_masks[mask, :, :]
pred_ranks = pred_ranks[mask]
elif dataset == 'irsr':
if len(pred_ranks) > 8:
sorted_data = sorted(enumerate(pred_ranks), key=lambda x: x[1], reverse=True)
top8_indices = [index for index, value in sorted_data[:8]]
mask = [i in top8_indices for i in range(len(pred_ranks))]
pred_masks = pred_masks[mask, :, :]
pred_ranks = pred_ranks[mask]
elif dataset == 'sifr':
if len(pred_ranks) > 41:
sorted_data = sorted(enumerate(pred_ranks), key=lambda x: x[1], reverse=True)
top41_indices = [index for index, value in sorted_data[:41]]
mask = [i in top41_indices for i in range(len(pred_ranks))]
pred_masks = pred_masks[mask, :, :]
pred_ranks = pred_ranks[mask]
res.append({'gt_masks': [mask for mask in gt_masks], 'segmaps': pred_masks, 'gt_ranks': gt_ranks,'rank_scores': [rank for rank in pred_ranks], 'img_name': name})
saliency_rank = [rank for rank in pred_ranks]
all_segmaps = np.zeros_like(gt_masks[0], dtype=np.float)
segmaps1 = copy.deepcopy(pred_masks)
if len(pred_masks) != 0:
color_index = [sorted(saliency_rank).index(a) + 1 for a in saliency_rank]
color_len = len(color_index)
if dataset == 'assr':
if color_len <= 10:
color = [math.floor(255. / 10 * (a + (10 - color_len))) for a in color_index]
else:
color = [max(math.floor(255. / 10 * (a + (10 - color_len))), 25) for a in color_index]
else:
color = [255. / len(saliency_rank) * a for a in color_index]
cover_region = all_segmaps != 0
for i in range(len(segmaps1), 0, -1):
obj_id_list = find_all_indexes(color_index, i)
if len(obj_id_list) == 0:
continue
else:
for obj_id in obj_id_list:
seg = segmaps1[obj_id]
seg[seg >= 0.5] = color[obj_id]
seg[seg < 0.5] = 0
seg[cover_region] = 0
all_segmaps += seg
cover_region = all_segmaps != 0
all_segmaps = all_segmaps.astype(np.int)
cv2.imwrite(saliencymapPath + '{}.png'.format(name[:-4]), all_segmaps)
else:
all_segmaps = np.zeros_like(gt_masks[0], dtype=np.int)
cv2.imwrite(saliencymapPath + '{}.png'.format(name[:-4]), all_segmaps)
segmapsforsasor = np.zeros([0, img_height, img_width])
print("Image:"+name+"------Pred_masks is None after confidence threshold")
res.append({'gt_masks': [mask for mask in gt_masks], 'segmaps': segmapsforsasor,'gt_ranks': [rank for rank in gt_ranks],'rank_scores': [], 'img_name': name})
if comm.is_main_process():
r_corre = rank_evalu(res, 0.5)
mae, avg_spr_norm, image_used=doOffcialSorInference(SOR_DATASETPATH,saliencymapPath,dataset,datasetmode)
f_m = mf_evalu(res)
with open(ourputdir+f"{model_name.split('.')[0]}_{datasetmode}metricsresult_Thres{cfg.EVALUATION.RESULT_THRESHOLD}.txt", "a") as f:
f.write('\n------------\n')
f.write(f'model_dir: {model_root_dir}\n')
f.write(f'model_name: {model_name}\n')
f.write("SA-SOR: %.4f\n" % r_corre)
f.write('Official SOR: {}\n'.format(avg_spr_norm))
f.write('Official Image Used: {}\n'.format(image_used))
f.write('MAE: {}\n'.format(mae))
f.write('F-measure: {}\n'.format(f_m))
return mae,avg_spr_norm,image_used,r_corre,f_m