MaskClustering/preprocess/scannet/prepare_single_gt.py

import os
import argparse
import json
import numpy as np
import pandas as pd
import sys
sys.path.append('/home/jovyan/users/bulat/workspace/3drec/MaskClustering')
print("prepare_single_gt sys.path", sys.path)
from evaluation.constants import SCANNET_IDS

def export_gt(filename, label_ids, instance_ids):
    gt_data = label_ids * 1000 + instance_ids + 1
    np.savetxt(filename, gt_data, fmt='%d')

def point_indices_from_group(seg_indices, group, labels_pd):
    group_segments = np.array(group['segments'])
    label = group['label']

    # Map the category name to id
    label_ids = labels_pd[labels_pd['raw_category'] == label]['id']
    label_id = int(label_ids.iloc[0]) if len(label_ids) > 0 else 0

    # Only store for the valid categories
    if not label_id in SCANNET_IDS:
        label_id = 0

    # get points, where segment indices (points labelled with segment ids) are in the group segment list
    point_IDs = np.where(np.isin(seg_indices, group_segments))

    return point_IDs[0], label_id

def handle_single_scene(scene_path, output_path, labels_pd, scene_name):
    segments_file = os.path.join(scene_path, f'{scene_name}_vh_clean_2.0.010000.segs.json')
    aggregations_file = os.path.join(scene_path, f'{scene_name}.aggregation.json')

    output_gt_file = os.path.join(output_path, f'{scene_name}.txt')
 
    # Load segments file
    with open(segments_file) as f:
        segments = json.load(f)
        seg_indices = np.array(segments['segIndices'])

    # Load Aggregations file
    with open(aggregations_file) as f:
        aggregation = json.load(f)
        seg_groups = np.array(aggregation['segGroups'])

    # Generate new labels
    labelled_pc = np.zeros((len(seg_indices), 1))
    instance_ids = np.zeros((len(seg_indices), 1))
    for group in seg_groups:
        p_inds, label_id = point_indices_from_group(seg_indices, group, labels_pd)

        labelled_pc[p_inds] = label_id
        instance_ids[p_inds] = group['id'] + 1

    labelled_pc = labelled_pc.astype(int)
    instance_ids = instance_ids.astype(int)

    export_gt(output_gt_file, labelled_pc, instance_ids)
    print(f"Ground truth сохранен в {output_gt_file}")

if __name__ == '__main__':
    parser = argparse.ArgumentParser()
    parser.add_argument('--scene_path', required=True, type=str, help='Путь к директории сцены')
    parser.add_argument('--gt_dir', required=True, type=str, help='Директория для сохранения ground truth')
    parser.add_argument('--label_map', required=True, type=str, help='Путь к файлу маппинга меток')
    parser.add_argument('--scene_name', required=True, type=str, help='Имя сцены')
    args = parser.parse_args()

    # Load label map
    labels_pd = pd.read_csv(args.label_map, sep='\t', header=0)
    
    os.makedirs(args.gt_dir, exist_ok=True)
    
    handle_single_scene(args.scene_path, args.gt_dir, labels_pd, args.scene_name)