TreeLearn/offset_calculation_test.py at main · MadPadCF/TreeLearn · 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
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
import torch
import os
import numpy as np
import argparse
import pprint
import shutil
from tree_learn.dataset import TreeDataset
from tree_learn.model import TreeLearn
from tree_learn.util import (build_dataloader, get_root_logger, load_checkpoint, ensemble,
                             get_coords_within_shape, get_hull_buffer, get_hull, get_cluster_means,
                             propagate_preds, save_treewise, load_data, save_data, make_labels_consecutive,
                             get_config, generate_tiles, assign_remaining_points_nearest_neighbor,
                             get_pointwise_preds, get_instances)

TREE_CLASS_IN_PYTORCH_DATASET = 0
NON_TREES_LABEL_IN_GROUPING = 0
NOT_ASSIGNED_LABEL_IN_GROUPING = -1
START_NUM_PREDS = 1


def run_treelearn_pipeline(config, config_path=None):
    # make dirs
    plot_name = os.path.basename(config.forest_path)[:-4]
    base_dir = os.path.dirname(os.path.dirname(config.forest_path))
    documentation_dir = os.path.join(base_dir, 'documentation')
    unvoxelized_data_dir = os.path.join(base_dir, 'forest')
    voxelized_data_dir = os.path.join(base_dir, f'forest_voxelized{config.sample_generation.voxel_size}')
    tiles_dir = os.path.join(base_dir, 'tiles')
    results_dir = os.path.join(base_dir, 'results')
    os.makedirs(documentation_dir, exist_ok=True)
    os.makedirs(unvoxelized_data_dir, exist_ok=True)
    os.makedirs(voxelized_data_dir, exist_ok=True)
    os.makedirs(tiles_dir, exist_ok=True)
    os.makedirs(results_dir, exist_ok=True)

    # documentation
    logger = get_root_logger(os.path.join(documentation_dir, 'log_pipeline'))
    logger.info(pprint.pformat(config, indent=2))
    if config_path is not None:
        shutil.copy(args.config, os.path.join(documentation_dir, os.path.basename(args.config)))

    # generate tiles used for inference and specify path to it in dataset config
    config.dataset_test.data_root = os.path.join(tiles_dir, 'npz')
    if config.tile_generation:
        logger.info('#################### generating tiles ####################')
        generate_tiles(config.sample_generation, config.forest_path, logger)

    # Make pointwise predictions with pretrained model
    logger.info(f'{plot_name}: #################### getting pointwise predictions ####################')
    model = TreeLearn(**config.model).cuda()
    dataset = TreeDataset(**config.dataset_test, logger=logger)
    dataloader = build_dataloader(dataset, training=False, **config.dataloader)

    #

    # print(f'Dataset getitem index 0: {dataset.__getitem__(0)}')
    xyz, input_feat, instance_label, semantic_label, pt_offset_label, center, mask_inner, mask_off, mask_sem = dataset.__getitem__(0)
    print(f'instance_label: {torch.unique(semantic_label)}')

    # load_checkpoint(config.pretrain, logger, model)
    # pointwise_results = get_pointwise_preds(model, dataloader, config.model, logger)
    # semantic_prediction_logits, semantic_labels, offset_predictions, offset_labels, coords, instance_labels, backbone_feats, input_feats = pointwise_results
    # del model

    # # ensemble predictions from overlapping tiles
    # logger.info(f'{plot_name}: #################### ensembling predictions ####################')
    # data = ensemble(coords, semantic_prediction_logits, semantic_labels, offset_predictions,
    #                 offset_labels, instance_labels, backbone_feats, input_feats)
    # coords, semantic_prediction_logits, semantic_labels, offset_predictions, offset_labels, instance_labels, backbone_feats, input_feats = data

    # # get mask of inner coords if outer points should be removed
    # if config.shape_cfg.outer_remove:
    #     logger.info(f'{plot_name}: #################### remove outer points ####################')
    #     hull_buffer_large = get_hull_buffer(coords, config.shape_cfg.alpha, buffersize=config.shape_cfg.outer_remove)
    #     mask_coords_within_hull_buffer_large = get_coords_within_shape(coords, hull_buffer_large)
    #     masks_inner_coords = np.logical_not(mask_coords_within_hull_buffer_large)

    # # get tree detections
    # logger.info(f'{plot_name}: #################### getting predicted instances ####################')
    # instance_preds = get_instances(coords, offset_predictions, semantic_prediction_logits, config.grouping, input_feats[:, -1], TREE_CLASS_IN_PYTORCH_DATASET, NON_TREES_LABEL_IN_GROUPING, NOT_ASSIGNED_LABEL_IN_GROUPING, START_NUM_PREDS)
    # instance_preds_after_initial_clustering = np.copy(instance_preds)

    # # assign remaining points
    # tree_mask = instance_preds != NON_TREES_LABEL_IN_GROUPING
    # instance_preds[tree_mask] = assign_remaining_points_nearest_neighbor(coords[tree_mask] + offset_predictions[tree_mask], instance_preds[tree_mask], NOT_ASSIGNED_LABEL_IN_GROUPING)

    # # save pointwise results
    # if config.save_cfg.save_pointwise:
    #     pointwise_dir = os.path.join(results_dir, 'pointwise_results')
    #     os.makedirs(pointwise_dir, exist_ok=True)
    #     pointwise_results = {
    #         'coords': coords,
    #         'offset_predictions': offset_predictions,
    #         'offset_labels': offset_labels,
    #         'semantic_prediction_logits': semantic_prediction_logits,
    #         'semantic_labels': semantic_labels,
    #         'instance_labels': instance_labels,
    #         'backbone_feats': backbone_feats,
    #         'input_feats': input_feats,
    #         'instance_preds': instance_preds,
    #         'instance_preds_after_initial_clustering': instance_preds_after_initial_clustering
    #     }
    #     if config.shape_cfg.outer_remove:
    #         pointwise_results['masks_inner_coords'] = masks_inner_coords
    #         hull_buffer_large.to_pickle(os.path.join(pointwise_dir, 'hull_buffer_large.pkl'))

    #     np.savez_compressed(os.path.join(pointwise_dir, 'pointwise_results.npz'), **pointwise_results)

    # # remove outer points with buffer
    # if config.shape_cfg.outer_remove:
    #     coords, semantic_prediction_logits, semantic_labels, offset_predictions, offset_labels, instance_labels, instance_preds = \
    #         coords[masks_inner_coords], semantic_prediction_logits[masks_inner_coords], \
    #         semantic_labels[masks_inner_coords], offset_predictions[masks_inner_coords], \
    #         offset_labels[masks_inner_coords], instance_labels[masks_inner_coords], \
    #         instance_preds[masks_inner_coords]
    #     instance_preds[instance_preds != NON_TREES_LABEL_IN_GROUPING] = make_labels_consecutive(instance_preds[instance_preds != NON_TREES_LABEL_IN_GROUPING], start_num=1)

    # # get information whether tree clusters are within or outside hull (used for saving tree in different categories later)
    # if config.save_cfg.save_treewise:
    #     cluster_means = get_cluster_means(coords[instance_preds != NON_TREES_LABEL_IN_GROUPING] + offset_predictions[instance_preds != NON_TREES_LABEL_IN_GROUPING],
    #                                       instance_preds[instance_preds != NON_TREES_LABEL_IN_GROUPING])
    #     hull = get_hull(coords[:, :2], config.shape_cfg.alpha)
    #     cluster_means_within_hull = get_coords_within_shape(cluster_means, hull)

    #     # get information whether trees have points very close to hull (used for saving trees in different categories later)
    #     hull_buffer_small = get_hull_buffer(coords, config.shape_cfg.alpha, buffersize=config.shape_cfg.buffer_size_to_determine_edge_trees)
    #     mask_coords_at_edge = get_coords_within_shape(coords, hull_buffer_small)
    #     instance_preds_at_edge = np.unique(instance_preds[mask_coords_at_edge])
    #     instance_preds_at_edge = np.delete(instance_preds_at_edge, np.where(instance_preds_at_edge == NON_TREES_LABEL_IN_GROUPING))
    #     insts_not_at_edge = np.ones(len(cluster_means_within_hull))
    #     insts_not_at_edge[instance_preds_at_edge-1] = 0
    #     insts_not_at_edge = insts_not_at_edge.astype('bool')

    # # propagate predictions to original forest
    # if config.save_cfg.return_type == 'original':
    #     logger.info(f'{plot_name}: Propagating predictions to original points')
    #     coords_to_return = load_data(config.forest_path)[:, :3]
    #     if config.shape_cfg.outer_remove:
    #         mask_coords_to_return_within_hull_buffer_large = get_coords_within_shape(coords_to_return, hull_buffer_large)
    #         masks_inner_coords_to_return = np.logical_not(mask_coords_to_return_within_hull_buffer_large)
    #         coords_to_return = coords_to_return[masks_inner_coords_to_return]
    #     preds_to_return = propagate_preds(coords, instance_preds, coords_to_return, n_neighbors=5)
    # elif config.save_cfg.return_type == 'voxelized':
    #     logger.info(f'{plot_name}: Propagating predictions to voxelized points')
    #     voxelized_forest_path = os.path.join(voxelized_data_dir, f'{plot_name}.npz')
    #     coords_to_return = load_data(voxelized_forest_path)[:, :3]
    #     if config.shape_cfg.outer_remove:
    #         mask_coords_to_return_within_hull_buffer_large = get_coords_within_shape(coords_to_return, hull_buffer_large)
    #         masks_inner_coords_to_return = np.logical_not(mask_coords_to_return_within_hull_buffer_large)
    #         coords_to_return = coords_to_return[masks_inner_coords_to_return]
    #     preds_to_return = propagate_preds(coords, instance_preds, coords_to_return, n_neighbors=5)
    # elif config.save_cfg.return_type == 'voxelized_and_filtered':
    #     coords_to_return = coords
    #     preds_to_return = instance_preds

    # # save
    # logger.info(f'{plot_name}: #################### Saving ####################')
    # full_dir = os.path.join(results_dir, 'full_forest')
    # os.makedirs(full_dir, exist_ok=True)

    # for save_format in config.save_cfg.save_formats:
    #     save_data(np.hstack([coords_to_return, preds_to_return.reshape(-1, 1)]), save_format, plot_name, full_dir)
    #     save_data(np.hstack([coords_to_return, preds_to_return.reshape(-1, 1)]), save_format, plot_name, full_dir)
    # if config.save_cfg.save_treewise:
    #     trees_dir = os.path.join(results_dir, 'individual_trees')
    #     os.makedirs(trees_dir, exist_ok=True)
    #     save_treewise(coords_to_return, preds_to_return, cluster_means_within_hull, insts_not_at_edge, "las", trees_dir, NON_TREES_LABEL_IN_GROUPING)
    # return


if __name__ == '__main__':
    parser = argparse.ArgumentParser('tree_learn')
    parser.add_argument('--config', type=str, help='path to config file for pipeline')
    args = parser.parse_args()
    config = get_config(args.config)
    run_treelearn_pipeline(config, args.config)