-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathanalyse.py
180 lines (146 loc) · 7.47 KB
/
analyse.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
import torch
import torch.nn as nn
import numpy as np
# # Create a random tensor to represent your input
# input = torch.rand([1, 1536, 24, 42])
# # Define the layer
# layer = nn.ModuleList([
# # nn.ConvTranspose2d(in_channels=1536, out_channels=1536, kernel_size=3, stride=2, padding=1, dilation=1, output_padding=1),
# nn.Conv2d(1536,256 , kernel_size=7, stride=1, padding=3, bias=False),
# nn.ReLU(inplace=True),
# nn.UpsamplingNearest2d(scale_factor=2), # x2
# nn.Conv2d(256, 64, kernel_size=7, stride=1, padding=3, bias=False),
# nn.ReLU(inplace=True),
# nn.UpsamplingNearest2d(scale_factor=2), # x2
# # nn.Conv2d(64, 64, kernel_size=7, stride=1, padding=3, bias=False),
# # nn.ReLU(inplace=True),
# nn.ConvTranspose2d(64,64, kernel_size=3, stride=2, padding=1, dilation=1, output_padding=1),
# nn.ReLU(inplace=True),
# nn.UpsamplingNearest2d(scale_factor=2), # x2
# # nn.Conv2d(64, 64, kernel_size=7, stride=1, padding=3, bias=False),
# nn.Conv2d(64, 1536, kernel_size=3, stride=1, padding=1, bias=False),
# nn.ReLU(inplace=True),
# nn.UpsamplingBilinear2d(scale_factor=2)
# # nn.ConvTranspose2d(64, 1536, kernel_size=3, stride=2, padding=1, dilation=1, output_padding=1),
# # nn.ReLU(inplace=True),
# # nn.ConvTranspose2d(64, 1536, kernel_size=3, stride=2, padding=1, dilation=1, output_padding=1),
# # nn.ReLU(inplace=True),
# # nn.UpsamplingNearest2d(scale_factor=4),
# ])
# # Pass the input through the layer
# output = input
# for fc in layer:
# output = fc(output)
# # Print the output size
# print(output.shape)
# # data = np.load('/home/coisini/data/nuscenes_fcclip_features/samples/CAM_BACK/n008-2018-08-01-15-16-36-0400__CAM_BACK__1533151603537558.npy')
# # print(data)
# # print(data.shape)
# import numpy as np
# res_voxel_coors = np.array([[1, 2, 3], [4, 5, 6], [4, 5, 6]]) # 假设这是你的原始数据
# i = 7 # 要填充的值
# # 在最后一维（列）后面添加一列，填充值为i
# res_voxel_coors = np.pad(res_voxel_coors, ((0, 0), (1, 0)), constant_values=i)
# print(res_voxel_coors)
# point_fcclip_features = torch.rand(6, 34720, 1536)
# pol_voxel_ind = torch.randint(0, 10000, size=(34720, 3))
# # 步骤1: 获取唯一的 voxel 索引和对应的计数
# unique_voxel_idx, inverse_idx, voxel_counts = torch.unique(pol_voxel_ind[:, 0], return_inverse=True, return_counts=True)
# num_voxels = len(unique_voxel_idx)
# print(pol_voxel_ind[:, 0])
# # 步骤2: 计算六个摄像头 point_fcclip_features 的平均值
# avg_point_fcclip_features = point_fcclip_features.mean(dim=0)
# print(avg_point_fcclip_features.shape)
# # 步骤3: 创建 voxel_fcclip_features 张量并填充数据
# voxel_fcclip_features = torch.zeros(num_voxels, 1536, device=point_fcclip_features.device)
# voxel_fcclip_features.index_add_(0, inverse_idx, avg_point_fcclip_features)
# voxel_fcclip_features /= voxel_counts.unsqueeze(1)
# print(voxel_fcclip_features.shape)
# gt_classes = [torch.zeros(size=[0])]
# gt_masks = [torch.zeros(size=[0,7938])]
# checkpoint = torch.load('/home/coisini/project/fc-clip/checkpoints/fcclip_cocopan.pth')
# print(checkpoint['model'].keys())
# import open_clip
# print(open_clip.list_pretrained())
# import clip
# print(clip.available_models())
# import os
# print(os.path.basename("/home/coisini/data/nuscenes_openseg_features/samples/LIDAR_TOP/n015-2018-07-24-11-22-45+0800__LIDAR_TOP__1532402927647951"))
# print(os.path.dirname("/home/coisini/data/nuscenes_openseg_features/samples/LIDAR_TOP/n015-2018-07-24-11-22-45+0800__LIDAR_TOP__1532402927647951.npy"))
# preds = torch.rand([133,13])
# targes = torch.randint(low=0,high=12,size=[133])
# weights = torch.rand([13,13])
# from mmdet.models.losses import cross_entropy
# print(cross_entropy(preds,targes))
NUSCENES_LABELS_16 = ['barrier', 'bicycle', 'bus', 'car', 'construction vehicle', 'motorcycle', 'person', 'traffic cone',
'trailer', 'truck', 'drivable surface', 'other flat', 'sidewalk', 'terrain', 'manmade', 'vegetation']
NUSCENES_LABELS_DETAILS = ['barrier', 'barricade', 'bicycle', 'bus', 'car', 'bulldozer', 'excavator', 'concrete mixer', 'crane', 'dump truck',
'motorcycle', 'person', 'pedestrian','traffic cone', 'trailer', 'semi trailer', 'cargo container', 'shipping container', 'freight container',
'truck', 'road', 'curb', 'traffic island', 'traffic median', 'sidewalk', 'grass', 'grassland', 'lawn', 'meadow', 'turf', 'sod',
'building', 'wall', 'pole', 'awning', 'tree', 'trunk', 'tree trunk', 'bush', 'shrub', 'plant', 'flower', 'woods']
import yaml
with open("nuscenes.yaml", 'r') as stream:
nuscenesyaml = yaml.safe_load(stream)
base_thing_list = [cl for cl, is_thing in nuscenesyaml['base_thing_class'].items() if is_thing]
base_stuff_list = [cl for cl, is_stuff in nuscenesyaml['base_stuff_class'].items() if is_stuff]
novel_thing_list = [cl for cl, is_thing in nuscenesyaml['novel_thing_class'].items() if is_thing]
novel_stuff_list = [cl for cl, is_stuff in nuscenesyaml['novel_stuff_class'].items() if is_stuff]
MAPPING_NUSCENES_DETAILS = np.array([0, 0, 1, 2, 3, 4, 4, 4, 4, 4,
5, 6, 6, 7, 8, 8, 8, 8, 8,
9, 10, 11, 11, 11, 12, 13, 13, 13, 13, 13, 13,
14, 14, 14, 14, 15, 15, 15, 15, 15, 15, 15, 15])+1
base_thing = []
base_stuff = []
novel_thing = []
novel_stuff = []
mapping_dict = dict()
for i in range(len(NUSCENES_LABELS_DETAILS)):
label = NUSCENES_LABELS_DETAILS[i]
id = MAPPING_NUSCENES_DETAILS[i]
mapping_dict[label] = id
for k in mapping_dict.keys():
if mapping_dict[k] in base_thing_list:
base_thing.append(k)
if mapping_dict[k] in base_stuff_list:
base_stuff.append(k)
if mapping_dict[k] in novel_thing_list:
novel_thing.append(k)
if mapping_dict[k] in novel_stuff_list:
novel_stuff.append(k)
base_total = base_thing+base_stuff
novel_total = novel_thing+novel_stuff
total = base_total+novel_total
base_total_add_noise = ['noise'] + base_total
total_add_noise = ['noise'] + total
# print(len(NUSCENES_LABELS_DETAILS))
# print(MAPPING_NUSCENES_DETAILS+1)
print(base_thing)
print(base_stuff)
print(novel_thing)
print(novel_stuff)
print(base_total,len(base_total))
print(total,len(total))
transfer_dict = list(np.vectorize(mapping_dict.__getitem__)(total))
transfer_dict = transfer_dict
transfer_list = []
count = {}
for num in transfer_dict:
if num not in count:
count[num] = len(count) + 1
transfer_list.append(count[num])
transfer_list = [0]+transfer_list
print(transfer_list)
print(transfer_dict)
print(mapping_dict)
import torch
# 假设你的class_scores是一个133x32的Tensor
class_scores = torch.randn(133, 32)
# 这是你的多标签索引
labels = [0, 1, 1, 2, 3, 4, 4, 4, 4, 4, 5, 6, 6, 6, 6, 6, 7, 8, 9, 9, 9, 10, 11, 11, 11, 11, 11, 11, 12, 12, 12, 12]
num_classes = max(labels) + 1 # 计算总类别数
# 创建一个新的Tensor，用于存储映射后的得分
class_preds = torch.zeros(133, num_classes)
# 对于每个类别，找到对应的标签，然后获取最大得分
for i, class_idx in enumerate(labels):
class_preds[:, class_idx], _ = torch.max(torch.stack([class_scores[:, i], class_preds[:, class_idx]], dim=-1), dim=-1)
print(class_preds.size()) # 输出新的class_score