-
Notifications
You must be signed in to change notification settings - Fork 165
/
Val_model_heatmap.py
207 lines (160 loc) · 6.42 KB
/
Val_model_heatmap.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
"""This is the main validation interface using heatmap trick
Author: You-Yi Jau, Rui Zhu
Date: 2019/12/12
"""
import numpy as np
import torch
from torch.autograd import Variable
import torch.backends.cudnn as cudnn
import torch.optim
import torch.nn as nn
import torch.nn.functional as F
import torch.utils.data
from tqdm import tqdm
from utils.loader import dataLoader, modelLoader, pretrainedLoader
import logging
from utils.tools import dict_update
from utils.utils import labels2Dto3D, flattenDetection, labels2Dto3D_flattened
from utils.utils import pltImshow, saveImg
from utils.utils import precisionRecall_torch
from utils.utils import save_checkpoint
from pathlib import Path
from models.model_wrap import SuperPointFrontend_torch
@torch.no_grad()
class Val_model_heatmap(SuperPointFrontend_torch):
def __init__(self, config, device='cpu', verbose=False):
self.config = config
self.model = self.config['name']
self.params = self.config['params']
self.weights_path = self.config['pretrained']
self.device=device
## other parameters
# self.name = 'SuperPoint'
# self.cuda = cuda
self.nms_dist = self.config['nms']
self.conf_thresh = self.config['detection_threshold']
self.nn_thresh = self.config['nn_thresh'] # L2 descriptor distance for good match.
self.cell = 8 # deprecated
self.cell_size = 8 # Size of each output cell. Keep this fixed.
self.border_remove = 4 # Remove points this close to the border.
self.sparsemap = None
self.heatmap = None # np[batch, 1, H, W]
self.pts = None
self.pts_subpixel = None
## new variables
self.pts_nms_batch = None
self.desc_sparse_batch = None
self.patches = None
pass
def loadModel(self):
# model = 'SuperPointNet'
# params = self.config['model']['subpixel']['params']
from utils.loader import modelLoader
self.net = modelLoader(model=self.model, **self.params)
checkpoint = torch.load(self.weights_path,
map_location=lambda storage, loc: storage)
self.net.load_state_dict(checkpoint['model_state_dict'])
self.net = self.net.to(self.device)
logging.info('successfully load pretrained model from: %s', self.weights_path)
pass
def extract_patches(self, label_idx, img):
"""
input:
label_idx: tensor [N, 4]: (batch, 0, y, x)
img: tensor [batch, channel(1), H, W]
"""
from utils.losses import extract_patches
patch_size = self.config['params']['patch_size']
patches = extract_patches(label_idx.to(self.device), img.to(self.device),
patch_size=patch_size)
return patches
pass
def run(self, images):
"""
input:
images: tensor[batch(1), 1, H, W]
"""
from Train_model_heatmap import Train_model_heatmap
from utils.var_dim import toNumpy
train_agent = Train_model_heatmap
with torch.no_grad():
outs = self.net(images)
semi = outs['semi']
self.outs = outs
channel = semi.shape[1]
if channel == 64:
heatmap = train_agent.flatten_64to1(semi, cell_size=self.cell_size)
elif channel == 65:
heatmap = flattenDetection(semi, tensor=True)
heatmap_np = toNumpy(heatmap)
self.heatmap = heatmap_np
return self.heatmap
pass
def heatmap_to_pts(self):
heatmap_np = self.heatmap
pts_nms_batch = [self.getPtsFromHeatmap(h) for h in heatmap_np] # [batch, H, W]
self.pts_nms_batch = pts_nms_batch
return pts_nms_batch
# def soft_argmax_points(self):
# """
# # make sure you have points ahead
# inputs:
# """
# # from utils.losses import extract_patches
# from utils.losses import extract_patch_from_points
# ##### check not take care of batch #####
# print("not take care of batch! only take first element!")
# pts = self.pts_nms_batch
# pts = pts[0].transpose().copy()
# patches = extract_patch_from_points(self.heatmap, pts, patch_size=5)
# import torch
# patches = np.stack(patches)
# patches_torch = torch.tensor(patches, dtype=torch.float32).unsqueeze(0)
# print("patches: ", patches_torch.shape)
# print("pts: ", pts.shape)
# dxdy = soft_argmax_2d(patches_torch)
# print("dxdy: ", dxdy.shape)
# points = pts
# points[:,:2] += dxdy.numpy().squeeze()
# self.pts_subpixel = [points.transpose().copy()]
# return self.pts_subpixel.copy()
# pass
def desc_to_sparseDesc(self):
# pts_nms_batch = [self.getPtsFromHeatmap(h) for h in heatmap_np]
desc_sparse_batch = [self.sample_desc_from_points(self.outs['desc'], pts) for pts in self.pts_nms_batch]
self.desc_sparse_batch = desc_sparse_batch
return desc_sparse_batch
if __name__ == '__main__':
# filename = 'configs/magicpoint_shapes_subpix.yaml'
filename = 'configs/magicpoint_repeatability_heatmap.yaml'
import yaml
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
torch.set_default_tensor_type(torch.FloatTensor)
with open(filename, 'r') as f:
config = yaml.load(f)
task = config['data']['dataset']
# data loading
from utils.loader import dataLoader_test as dataLoader
data = dataLoader(config, dataset='hpatches')
test_set, test_loader = data['test_set'], data['test_loader']
# load frontend
val_agent = Val_model_heatmap(config['model'], device=device)
# take one sample
for i, sample in tqdm(enumerate(test_loader)):
if i>1: break
val_agent.loadModel()
# points from heatmap
img = sample['image']
print("image: ", img.shape)
heatmap_batch = val_agent.run(img.to(device)) # heatmap: numpy [batch, 1, H, W]
# heatmap to pts
pts = val_agent.heatmap_to_pts()
# print("pts: ", pts)
print("pts[0]: ", pts[0].shape)
print("pts: ", pts[0][:,:3])
pts_subpixel = val_agent.soft_argmax_points(pts)
print("subpixels: ", pts_subpixel[0][:,:3])
# heatmap, pts to desc
desc_sparse = val_agent.desc_to_sparseDesc()
print("desc_sparse[0]: ", desc_sparse[0].shape)
# pts, desc, _, heatmap