# Copyright 2026 Toyota Research Institute. All rights reserved. import os import cv2 import numpy as np from glob import glob from anydata.utils.geometry import invert_extrinsics from anydata.utils.read import read_numpy, read_yaml, read_image, read_depth, read_json from anydata.utils.write import write_image, write_npz, write_json from anydata.converters.utils import run, add_key_to_dict, fill_metadata, parse_dst_seq, frame_name ####################################################### def convert_extrinsics(extrinsics): flip_yz = np.eye(4) flip_yz[1, 1] = -1 flip_yz[2, 2] = -1 extrinsics = np.matmul(extrinsics, flip_yz) return extrinsics def get_extrinsics(filename, transforms): # filename = filename.replace('images', 'poses').replace('.png', '.json') # extrinsics = np.array(read_json(filename)['pose']) # # extrinsics = convert_extrinsics(extrinsics) # # extrinsics = invert_extrinsics(extrinsics) # return extrinsics for transform in transforms['frames']: if transform['file_path'].endswith(os.path.basename(filename)): extrinsics = transform['transform_matrix'] extrinsics = convert_extrinsics(extrinsics) extrinsics = invert_extrinsics(extrinsics) return extrinsics return None ####################################################### def get_sequences(args): seqs = glob(f'{args.src}/**/transforms.json', recursive=True) seqs = [os.path.dirname(seq) for seq in seqs] return seqs def parse_sequence(seq, args): return parse_dst_seq(seq, args) ####################################################### def process_sequence(i, seq, dst, args): ### Initialize lists and dicts cameras = ['0'] num_frames = {cam: dict() for cam in cameras} resolution = {cam: dict() for cam in cameras} labels, lowdim = [], {} ### Get filenames filename_rgbs = sorted(glob(f'{seq}/images/*.jpg')) filename_poses = sorted(glob(f'{seq}/poses/*.json')) ### Read transforms for the sequence transforms = read_json(f'{seq}/transforms.json') ### Get intrinsics focal, cx, cy = transforms["fl_x"], transforms["cx"], transforms["cy"] intrinsics = np.array([[focal, 0, cx], [0, focal, cy], [0, 0, 1]]) ############ LOOP OVER CAMERAS for cam in cameras: ######## RGB FILENAMES for i, filename_rgb in enumerate(filename_rgbs): frame = frame_name(filename_rgb) ######## RGB if 'rgb' not in labels: labels.append('rgb') rgb = np.array(read_image(filename_rgb)) filename_rgb_out = f'{dst}/rgb/{cam}/{frame}.jpg' write_image(filename_rgb_out, rgb) ######## LOWDIM RGB filename_lowdim = add_key_to_dict(lowdim, f'{dst}/lowdim/{cam}/{frame}.npz') lowdim[filename_lowdim]['camera'] = cam lowdim[filename_lowdim]['timestep'] = int(frame) ### Get extrinsics extrinsics = get_extrinsics(filename_rgb, transforms) ######## INTRINSICS + EXTRINSICS if 'intrinsics' not in labels: labels.append('intrinsics') if 'extrinsics' not in labels: labels.append('extrinsics') filename_lowdim = add_key_to_dict(lowdim, f'{dst}/lowdim/{cam}/{frame}.npz') lowdim[filename_lowdim]['extrinsics'] = extrinsics lowdim[filename_lowdim]['intrinsics'] = intrinsics if len(filename_rgbs) > 0: resolution[cam]['rgb'] = rgb.shape[:2] num_frames[cam]['rgb'] = len(filename_rgbs) ######## WRITE LOWDIM for key, val in lowdim.items(): write_npz(key, val) ############ METADATA filename = f'{dst}/metadata.json' seq_metadata = fill_metadata( args=args, info=dict( name='Front3D', tags=['real','static','indoor'], raw_id=seq.replace(f'{args.src}/', ''), ), labels=labels, cameras=cameras, resolution=resolution, num_frames=num_frames, framerate=10, rgb=dict(extension='jpg'), intrinsics=dict(model='pinhole'), extrinsics=dict(transform='cam2world',metric=True), depth=dict(extension='npz',metric=True,sparse=True), semantic=None, action=None, language=None, specific=None, ) write_json(filename, seq_metadata) return dst ####################################################### if __name__ == '__main__': converter = os.path.basename(__file__) run(converter, get_sequences, parse_sequence, process_sequence) ####################################################### # import json # import numpy as np # from glob import glob # from datasets.dataloaders.BaseDataset import BaseDataset # from datasets.dataloaders.utils.FolderTree import FolderTree, merge_trees # from datasets.dataloaders.utils.misc import update_dict, invert_pose # from datasets.utils.read import read_image, read_depth # from vidar.utils.data import shuffle_dict # import os # import h5py # class HM3DDataset(BaseDataset): # def __init__(self, mask_depth=False, **kwargs): # super().__init__(**kwargs, base_tag='folder') # self.rgb_folder = 'images' # self.rgb_extension = '.png' # rgb_trees = [] # print("self.path", self.path) # folders = sorted(glob(f'{self.path}/*')) # print(folders) # for folder in folders: # rgb_trees.append(FolderTree( # f'{folder}', # context=self.context, # sub_folders=self.rgb_folder, # suffix=self.rgb_extension, # )) # self.rgb_tree = merge_trees(rgb_trees) # print(self.rgb_tree) # def __len__(self): # """Dataset length""" # return len(self.rgb_tree) # @staticmethod # def get_rgb(filename): # return read_image(filename) # def convert_pose(self,C2W): # flip_yz = np.eye(4) # flip_yz[1, 1] = -1 # flip_yz[2, 2] = -1 # C2W = np.matmul(C2W, flip_yz) # return C2W # def get_pose_and_K(self, filename): # parent_folder = os.path.dirname(os.path.dirname(os.path.dirname(filename))) # pose_filename = os.path.join(parent_folder, 'train', 'transforms.json') # poses = json.load(open(pose_filename)) # splits = filename.split('/') # image_name = splits[-2] + '/' + splits[-1] # transform_entry = next(frame for frame in poses['frames'] if frame['file_path'] == image_name) # # Extract the transform_matrix # C2W = transform_entry['transform_matrix'] # C2W = self.convert_pose(C2W) # #convert_pose to opencv # pose = invert_pose(C2W) # focal = poses["fl_x"] # cx = poses["cx"] # cy = poses["cy"] # intrinsics = np.array([[focal, 0, cx], [0, focal, cy], [0, 0, 1]]) # return pose, intrinsics # def add_target(self, sample, filename, time_cam, **kwargs): # update_dict(sample, 'filename', time_cam, filename) # pose, intrinsics = self.get_pose_and_K(filename) # if self.with_rgb: # update_dict(sample, 'rgb', time_cam, # self.get_rgb(filename)) # if self.with_intrinsics: # update_dict(sample, 'intrinsics', time_cam, # intrinsics) # if self.with_pose: # update_dict(sample, 'pose', time_cam, # pose) # self.add_dummy_data(sample, time_cam) # return sample # def add_context(self, sample, filename_context, **kwargs): # for time_cam, filename in filename_context.items(): # pose, intrinsics = self.get_pose_and_K(filename) # update_dict(sample, 'filename', time_cam, filename) # if self.with_rgb_context: # update_dict(sample, 'rgb', time_cam, # self.get_rgb(filename)) # if self.with_intrinsics_context: # update_dict(sample, 'intrinsics', time_cam, # intrinsics) # if self.with_pose_context: # update_dict(sample, 'pose', time_cam, # pose) # self.add_dummy_data_context(sample, time_cam) # return sample