-
Notifications
You must be signed in to change notification settings - Fork 3
/
evaluation_img.py
255 lines (217 loc) · 8.38 KB
/
evaluation_img.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
import argparse
import os
from typing import Dict, Tuple
import numpy as np
import torch
from torch.nn.functional import normalize
from torch.utils.data import DataLoader, Dataset
from torch.utils.data import Subset
import tqdm
def get_dataset(dataset_name: str, transformer=None) -> Dict:
# this is the root of the dataset
# Note:
# We have extracted the region feature for each image
# and saved it in the same location as the image,
# named "<ImageFileName>.npz".
# The region feature is for sgraf model in the paper
root = ""
if dataset_name == "sop":
from dataset_evalimg import SOP
trainset = SOP(root, "train",transform=transformer)
testset = SOP(root, "eval",transform=transformer)
trainset.num_classes = trainset.nb_classes()
return {"train": trainset, "test": testset, "metric": "rank1"}
elif dataset_name == "cub":
from dataset_evalimg import CUBirds
trainset = CUBirds(root, "train",transform=transformer)
testset = CUBirds(root, "eval",transform=transformer)
trainset.num_classes = trainset.nb_classes()
return {"train": trainset, "test": testset, "metric": "rank1"}
elif dataset_name == "car":
from dataset_evalimg import Cars
trainset = Cars(root, "train",transform=transformer)
testset = Cars(root, "eval",transform=transformer)
trainset.num_classes = trainset.nb_classes()
return {"train": trainset, "test": testset, "metric": "rank1"}
elif dataset_name == "inshop":
from dataset_evalimg import Inshop_Dataset
trainset = Inshop_Dataset(root, "train",transform=transformer)
query = Inshop_Dataset(root, "query",transform=transformer)
gallery = Inshop_Dataset(root, "gallery",transform=transformer)
trainset.num_classes = trainset.nb_classes()
return {"train": trainset, "query": query, "gallery": gallery, "metric": "rank1"}
elif dataset_name == "inat":
from dataset_evalimg import inaturalist
trainset = inaturalist.get_trainset(root, transform=transformer)
testset = inaturalist.get_testset(root, transform=transformer)
trainset.num_classes = 5690
return {"train": trainset, "test": testset, "metric": "rank1"}
else:
raise
@torch.no_grad()
def extract_feat(
model: torch.nn.Module,
dataset: Dataset,
batch_size: int,
num_workers: int) -> Tuple[torch.Tensor, torch.Tensor]:
n_data = len(dataset)
idx_all_rank = list(range(n_data))
dataset_this_rank = Subset(dataset, idx_all_rank)
kwargs = {
"batch_size": batch_size,
"num_workers": num_workers,
"drop_last": False
}
dataloader = DataLoader(dataset_this_rank, **kwargs)
x = None
y_np = []
idx = 0
for image, label in tqdm.tqdm(dataloader):
image = image.cuda()
embedding = get_feature(model, image)
embedding_size: int = embedding.size(1)
if x is None:
size = [len(idx_all_rank), embedding_size]
x = torch.zeros(*size, device=image.device)
x[idx:idx + embedding.size(0)] = embedding
y_np.append(np.array(label))
idx += embedding.size(0)
x = x.cpu()
y_np = np.concatenate(y_np, axis=0)
return x, y_np
@torch.no_grad()
def euclidean_distance(x, y, topk=2):
m, n = x.size(0), y.size(0)
xx = torch.pow(x, 2).sum(1, keepdim=True).expand(m, n)
yy = torch.pow(y, 2).sum(1, keepdim=True).expand(n, m).t()
dist = xx + yy
dist.addmm_(mat1=x, mat2=y.t(), beta=1, alpha=-2)
dist = dist.clamp(min=1e-12).sqrt()
return torch.topk(dist, topk, largest=False)
@torch.no_grad()
def get_metric(
query: torch.Tensor,
query_label: list,
gallery: torch.Tensor = None,
gallery_label: list = None,
l2norm=True,
metric="rank1"):
if gallery is None:
query = query.cuda()
if l2norm:
query = normalize(query)
query_label = query_label
list_pred = []
num_feat = query.size(0)
idx = 0
is_end = 0
while not is_end:
if idx + 128 < num_feat:
end = idx + 128
else:
end = num_feat
is_end = 1
_, index_pt = euclidean_distance(query[idx:end], query)
index_np = index_pt.cpu().numpy()[:, 1]
list_pred.append(index_np)
idx += 128
query_label = np.array(query_label).reshape(num_feat)
pred = np.concatenate(list_pred).reshape(num_feat)
rank_1 = np.sum(query_label == query_label[pred]) / num_feat
rank_1 = float(rank_1)
return rank_1 * 100
else:
query = query.cuda()
query_label = query_label
gallery = gallery.cuda()
gallery_label = np.array(gallery_label)
list_pred = []
if l2norm:
query = normalize(query)
gallery = normalize(gallery)
num_feat = query.size(0)
idx = 0
is_end = 0
while not is_end:
if idx + 128 < num_feat:
end = idx + 128
else:
end = num_feat
is_end = 1
_, index_pt = euclidean_distance(query[idx:end], gallery)
index_np = index_pt.cpu().numpy()[:, 0]
list_pred.append(index_np)
idx += 128
query_label = np.array(query_label).reshape(num_feat)
pred = np.concatenate(list_pred).reshape(num_feat)
rank_1 = np.sum(query_label == gallery_label[pred]) / num_feat
rank_1 = float(rank_1)
return rank_1 * 100
@torch.no_grad()
def evaluation(model: torch.nn.Module,
dataset_dict: Dict, batch_size: int, num_workers: int):
if "index" in dataset_dict:
raise NotImplementedError
elif "test" in dataset_dict:
dataset = dataset_dict["test"]
x, y = extract_feat(model, dataset, batch_size, num_workers)
metric = get_metric(x, y)
return metric
elif "query" in dataset_dict and "gallery" in dataset_dict:
dataset_q = dataset_dict["query"]
dataset_g = dataset_dict["gallery"]
q, q_label = extract_feat(model, dataset_q, batch_size, num_workers)
g, g_label = extract_feat(model, dataset_g, batch_size, num_workers)
metric = get_metric(query=q, query_label=q_label,
gallery=g, gallery_label=g_label)
return metric
def get_feature(model, image):
# return model.encode_image(image)
# return model.encode_image(image, cross_modal=True)
return model.encode_image(image, cross_modal=False)
def get_model(device, modelName):
from unire.model import unire
checkPointPath = modelName
print("Loading model: ", checkPointPath)
checkpoint = torch.load(checkPointPath, map_location='cpu')
state_dict = checkpoint['model']
args = argparse.Namespace()
args.gpu = torch.device(device)
model = unire(args, checkpoint['config'])
model.load_state_dict(state_dict)
model.eval()
model.to(device)
return model, model.preprocess
# ['RN50', 'RN101', 'RN50x4', 'RN50x16', 'RN50x64', 'ViT-B/32', 'ViT-B/16', 'ViT-L/14', 'ViT-L/14@336px']
def get_clip_model(device):
modelName = "ViT-L/14@336px"
print("Loading model: ", modelName)
from clip import clip
clip_model, preprocess = clip.load(modelName, device=device, jit=False)
clip_model.eval()
return clip_model, preprocess
if __name__ == '__main__':
processList = []
fileList = list(os.walk("output/vitb32/coco"))
for val in fileList:
if "checkpoint_best.pth" in val[2]:
processList.append(os.path.join(val[0], "checkpoint_best.pth"))
for modelName in processList:
datasetNameList = ["cub","sop", "inshop","inat"]
batch_size = 128
num_workers = 4
model, preprocess = get_model("cuda:0", modelName)
scores = []
for datasetName in datasetNameList:
print(datasetName)
dataset_dict: Dict = get_dataset(datasetName, preprocess)
score = evaluation(model, dataset_dict, batch_size, num_workers)
scores.append(score)
if isinstance(score, Tuple):
for i in score:
print(i, end=",")
else:
print(score, end=",")
print("\n")
print("scores: ", scores)
print("mean score: ", np.mean(scores))