File size: 5,936 Bytes
8d5039c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import glob
import tqdm
import os
import torch
import numpy as np
import open3d as o3d
from scipy.optimize import linear_sum_assignment
import itertools
from model.pointnet_util import *
from model.model_utils import *

def writePoints(points, clsRoad):
    with open(clsRoad, 'w+') as file1:
        for i in range(len(points)):
            point = points[i]
            file1.write(str(point[0]))
            file1.write(' ')
            file1.write(str(point[1]))
            file1.write(' ')
            file1.write(str(point[2]))
            file1.write('\n')

def writeEdges(edges, clsRoad):
    with open(clsRoad, 'w+') as file1:
        for i in range(len(edges)):
            edge = edges[i]
            file1.write(str(edge[0] + 1))
            file1.write(' ')
            file1.write(str(edge[1] + 1))
            file1.write(' ')
            file1.write('\n')

def assign_targets(points, gvs, radius):
    idx = ball_center_query(radius, points, gvs).type(torch.int64)
    batch_size = gvs.size()[0]
    idx_add = torch.arange(batch_size).to(idx.device).unsqueeze(-1).repeat(1, idx.shape[-1]) * gvs.shape[1]
    gvs = gvs.view(-1, 3)
    idx_add += idx
    target_points = gvs[idx_add.view(-1)].view(batch_size, -1, 3)
    dis = target_points - points
    dis[idx < 0] = 0
    dis /= radius
    label = torch.where(idx >= 0, torch.ones(idx.shape).to(idx.device), torch.zeros(idx.shape).to(idx.device))
    return dis, label

def test_model(model, data_loader, logger):
    dataloader_iter = iter(data_loader)
    with tqdm.trange(0, len(data_loader), desc='test', dynamic_ncols=True) as tbar:
        model.use_edge = True
        statistics = {'tp_pts': 0, 'num_label_pts': 0, 'num_pred_pts': 0, 'pts_bias': np.zeros(3, float),
                      'tp_edges': 0, 'num_label_edges': 0, 'num_pred_edges': 0}
        for cur_it in tbar:
            batch = next(dataloader_iter)
            load_data_to_gpu(batch)
            with torch.no_grad():
                batch = model(batch)
                load_data_to_cpu(batch)
                # print("test")
                keys = batch['frame_id']
                wf_vertices = batch['refined_keypoint']#.squeeze(0)
                wf_edges = batch['pair_points']#.squeeze(0)
                return keys, wf_vertices, wf_edges

# def test_model(model, data_loader, logger):
#     dataloader_iter = iter(data_loader)
#     with tqdm.trange(0, len(data_loader), desc='test', dynamic_ncols=True) as tbar:
#         model.use_edge = True
#         statistics = {'tp_pts': 0, 'num_label_pts': 0, 'num_pred_pts': 0, 'pts_bias': np.zeros(3, float),
#                       'tp_edges': 0, 'num_label_edges': 0, 'num_pred_edges': 0}
#         for cur_it in tbar:
#             batch = next(dataloader_iter)
#             load_data_to_gpu(batch)
#             with torch.no_grad():
#                 batch = model(batch)
#                 load_data_to_cpu(batch)
#             eval_process(batch, statistics)
#         bias = statistics['pts_bias'] / statistics['tp_pts']
#         logger.info('pts_recall: %f' % (statistics['tp_pts'] / statistics['num_label_pts']))
#         logger.info('pts_precision: %f' % (statistics['tp_pts'] / statistics['num_pred_pts']))
#         logger.info('pts_bias: %f, %f, %f' % (bias[0], bias[1], bias[2]))
#         logger.info('edge_recall: %f' % (statistics['tp_edges'] / statistics['num_label_edges']))
#         logger.info('edge_precision: %f' % (statistics['tp_edges'] / statistics['num_pred_edges']))

def eval_process(batch, statistics):
    batch_size = batch['batch_size']
    pts_pred, pts_refined, pts_label = batch['keypoint'], batch['refined_keypoint'], batch['vectors']
    edge_pred, edge_label = batch['edge_score'], batch['edges']
    mm_pts = batch['minMaxPt']
    id = batch['frame_id']

    idx = 0
    for i in range(batch_size):
        mm_pt = mm_pts[i]
        minPt = mm_pt[0]
        maxPt = mm_pt[1]
        deltaPt = maxPt - minPt

        p_pts = pts_refined[pts_pred[:, 0] == i]
        l_pts = pts_label[i]
        l_pts = l_pts[np.sum(l_pts, -1, keepdims=False) > -2e1]
        vec_a = np.sum(p_pts ** 2, -1)
        vec_b = np.sum(l_pts ** 2, -1)
        dist_matrix = vec_a.reshape(-1, 1) + vec_b.reshape(1, -1) - 2 * np.matmul(p_pts, np.transpose(l_pts))
        dist_matrix = np.sqrt(dist_matrix + 1e-6)
        p_ind, l_ind = linear_sum_assignment(dist_matrix)
        mask = dist_matrix[p_ind, l_ind] < 0.1   # 0.1
        tp_ind, tl_ind = p_ind[mask], l_ind[mask]
        #dis = np.abs(p_pts[tp_ind] - l_pts[tl_ind])
        dis = np.abs( ((p_pts[tp_ind]*deltaPt) + minPt) - ((l_pts[tl_ind]*deltaPt) + minPt) )


        statistics['tp_pts'] += tp_ind.shape[0]
        statistics['num_label_pts'] += l_pts.shape[0]
        statistics['num_pred_pts'] += p_pts.shape[0]
        statistics['pts_bias'] += np.sum(dis, 0)

        match_edge = list(itertools.combinations(l_ind, 2))
        match_edge = np.array([tuple(sorted(e)) for e in match_edge])
        score = edge_pred[idx:idx+len(match_edge)]
        idx += len(match_edge)
        l_edge = edge_label[i]
        l_edge = l_edge[np.sum(l_edge, -1, keepdims=False) > 0]
        l_edge = [tuple(e) for e in l_edge]
        match_edge = match_edge[score > 0.5]
        tp_edges = np.sum([tuple(e) in l_edge for e in match_edge])
        statistics['tp_edges'] += tp_edges
        statistics['num_label_edges'] += len(l_edge)
        statistics['num_pred_edges'] += match_edge.shape[0]


def load_data_to_gpu(batch_dict):
    device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
    for key, val in batch_dict.items():
        if not isinstance(val, np.ndarray):
            continue
        batch_dict[key] = torch.from_numpy(val).float().to(device)


def load_data_to_cpu(batch_dict):
    for key, val in batch_dict.items():
        if not isinstance(val, torch.Tensor):
            continue
        batch_dict[key] = val.cpu().numpy()