gnn_segmentation/pipeline.py at main · RePAIRProject/gnn_segmentation · GitHub

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
from network.gnns import detectionGCN, recognitionGCN
from utils.train_test_util import predict, show_results, add_noise
import os, sys
import random
import pickle
from torch_geometric.loader import DataLoader
import open3d as o3d
import numpy as np
import matplotlib.pyplot as plt
import torch
import torch_geometric.transforms as T
from torch_geometric.data import Data
from torch_geometric.loader import DataLoader
import yaml
from utils.integration import detect, torch_data_from_o3d_pcl, recognize_objects
import copy

if __name__ == '__main__':

    # dataset_root_path = '/media/lucap/big_data/datasets/repair/sand_detection_dataset_bb_yolo_1000scenes'
    # pcl = o3d.io.read_point_cloud(os.path.join(dataset_root_path, 'group_0015', 'pointclouds', f'PCL_000000.ply'))
    # xyz = np.asarray(pcl.points)
    # rgb = np.asarray(pcl.colors)
    # data_pts = test_set[0].x
    # np_pts = np.concatenate([xyz, rgb], axis=1)
    # breakpoint()
    # pts = torch.from_numpy(np_pts).type(torch.float32).to(device)
    print('reading config file..')
    cfg_file_path = os.path.join('configs', f'pipeline_cfg.yaml')
    with open(cfg_file_path, 'r') as yf:
        cfg = yaml.safe_load(yf)


    # DATASET
    dataset_root_path = cfg['dataset_root_path']
    #'/media/lucap/big_data/datasets/repair/sand_detection_dataset_bb_yolo_1000scenes'

    # CUDA DEVICE
    print('device')
    device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')

    # LOAD THE MODEL
    print("detection model")
    # MODELS
    detection_model_folder = cfg['detection_model_folder']
    # detection_model_path = os.path.join(detection_model_folder, 'model.pt')
    detection_best_weights = os.path.join(detection_model_folder, 'last.pth')
    parameters_det = {'input_features': 6, 'hidden_channels': 64, 'k': 10}
    detection_model = detectionGCN(input_features=parameters_det['input_features'], \
        hidden_channels=parameters_det['hidden_channels'])
    weights_dict = torch.load(detection_best_weights, map_location=device)
    detection_model.load_state_dict(state_dict=weights_dict)
    detection_model.eval()

    # if device == torch.device('cpu'):
    #     detection_model2 = torch.load(detection_model_path, map_location=torch.device('cpu'))
    # else:
    #     detection_model = torch.load(detection_model_path).to(device)
    # detection_model = torch.load(detection_model_path).to(device)
    # detection_model.load_state_dict(torch.load(detection_best_weights))

    # LOAD DATA
    # with open('/media/palma/D36D-688D/uni palma/gnn_segmentation/data/detection_sand_test_set', 'rb') as valid_set_file:
    #     validation_set = pickle.load(valid_set_file)
    # valid_loader = DataLoader(validation_set, batch_size=1, shuffle=True)
    # for data in valid_loader:  # Iterate in batches over the training dataset.
    #     # ADD NOISE
    #     data = add_noise(data, 0.5)
    #     data.x = data.x[:,:6]
    #     data.to(device)
    #     out = detection_model(data.x, data.edge_index)
    #     print("unique values", out.argmax(dim=1).unique())
    # breakpoint()

    # READ A RANDOM POINTCLOUD
    if len(sys.argv) > 1:
        group = sys.argv[1]
    else:
        group = '15'
    print(f"read random pointcloud of group {group}..")
    random_scene_idx = np.random.choice([995, 996, 997, 998, 999])
    pcl_init = o3d.io.read_point_cloud(os.path.join(dataset_root_path, f'group_00{group}', 'pointclouds', f'PCL_{random_scene_idx:06d}.ply'))
    o3d.visualization.draw_geometries([pcl_init], window_name='Input Pointcloud')

    # DETECT
    print('detecting..')
    detected_objects, background, preds = detect(pcl_init, detection_model, parameters_det, device)
    vis_detected_objects = copy.deepcopy(detected_objects)
    vis_detected_objects.paint_uniform_color([0,0,1])
    vis_background = copy.deepcopy(background)
    vis_background.paint_uniform_color([1,0,0])
    o3d.visualization.draw_geometries([vis_detected_objects, vis_background], window_name = 'detected objects (blue) and background (red)')

    print(f'recognition model for group {group}')
    if group == '15':
        g15_recognition_model_folder = cfg['g15_recognition_model_folder']
        # g15_recognition_model_path = os.path.join(g15_recognition_model_folder, 'model.pt')
        g15_recognition_model_weights = os.path.join(g15_recognition_model_folder, 'last.pth')
        parameters_rec_g15 = {'input_features': 6, 'hidden_channels': 64, 'output_classes': 17, 'k': 6}
        recognition_model_g15 = recognitionGCN(input_features=parameters_rec_g15['input_features'], \
            hidden_channels=parameters_rec_g15['hidden_channels'], output_classes=parameters_rec_g15['output_classes'])
        g15_weights_dict = torch.load(g15_recognition_model_weights, map_location=device)
        recognition_model_g15.load_state_dict(state_dict=g15_weights_dict)
        recognition_model_g15.eval()
        print('recognizing..')
        recognized_objects_as_pcls, pred_IDs = recognize_objects(detected_objects, recognition_model_g15, parameters_rec_g15, device)
    elif group == '29':
        g29_recognition_model_folder = cfg['g29_recognition_model_folder']
        g29_recognition_model_weights = os.path.join(g29_recognition_model_folder, 'last.pth')
        parameters_rec_g29 = {'input_features': 6, 'hidden_channels': 64, 'output_classes': 5, 'k': 6}
        recognition_model_g29 = recognitionGCN(input_features=parameters_rec_g29['input_features'], \
            hidden_channels=parameters_rec_g29['hidden_channels'], output_classes=parameters_rec_g29['output_classes'])
        g29_weights_dict = torch.load(g29_recognition_model_weights, map_location=device)
        recognition_model_g29.load_state_dict(state_dict=g29_weights_dict)
        recognition_model_g29.eval()
        print('recognizing..')
        recognized_objects_as_pcls, pred_IDs = recognize_objects(detected_objects, recognition_model_g29, parameters_rec_g29, device)


        # max_label = np.max(pred_IDs)
    # colors = plt.get_cmap("tab20")(pred_IDs / (max_label if max_label > 0 else 1))
    # recognized_objects_as_pcls.colors = o3d.utility.Vector3dVector(colors[:, :3])
    pcl_to_show = []
    bboxes = []
    if group == '15':
        cmap = plt.get_cmap('jet').resampled(17)
    elif group == '29':
        cmap = plt.get_cmap('jet').resampled(5)
    for rop, pid in zip(recognized_objects_as_pcls, pred_IDs):
        print('object with id', pid)
        col = cmap(pid)[:3]
        rop.paint_uniform_color(col)
        bbox = rop.get_axis_aligned_bounding_box()
        bbox.color = col
        pcl_to_show.append(rop)
        pcl_to_show.append(bbox)
        bboxes.append(bbox)

    o3d.visualization.draw_geometries(pcl_to_show, window_name = f'recognition on group {group} (objects)')
    o3d.visualization.draw_geometries([pcl_init]+bboxes, window_name = f'recognition on group {group} (scene)')
    breakpoint()