File size: 5,527 Bytes
319eb16
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
"""
EgoDex dataset loader for the generic dataset converter for Robometer model training.

This module provides a simple, readable loader inspired by the LIBERO loader:
- Discovers (HDF5, MP4) pairs per task directory
- Lazily loads frames via a small frame-loader callable
- Extracts a task description and pose actions from HDF5
- Returns a dictionary mapping task names to lists of trajectory dicts
"""

import os
from pathlib import Path
from re import A

import h5py
import numpy as np
from dataset_upload.helpers import generate_unique_id
from dataset_upload.video_helpers import load_video_frames
from tqdm import tqdm


class EgoDexFrameLoader:
    """Pickle-able frame loader for EgoDex videos."""

    def __init__(self, mp4_path: str):
        self.mp4_path = mp4_path

    def __call__(self) -> np.ndarray:
        """Load frames from the MP4 file when called."""
        return load_video_frames(Path(self.mp4_path), max_frames=1800)  # 30hz * 60s = 1800 frames


def _discover_trajectory_files(dataset_path: Path) -> list[tuple[Path, Path, str]]:
    """Discover all (HDF5, MP4) pairs grouped by task directory."""
    trajectory_files: list[tuple[Path, Path, str]] = []
    for task_dir in dataset_path.iterdir():
        if not task_dir.is_dir():
            continue
        task_name = task_dir.name
        for hdf5_file in task_dir.glob("*.hdf5"):
            mp4_file = hdf5_file.with_suffix(".mp4")
            if mp4_file.exists():
                trajectory_files.append((hdf5_file, mp4_file, task_name))
            else:
                print(f"Warning: Missing MP4 file for {hdf5_file}")
    return trajectory_files


def _load_hdf5_data(hdf5_path: Path) -> tuple[np.ndarray, str]:
    """Load pose data and task description from EgoDex HDF5 file."""
    with h5py.File(hdf5_path, "r") as f:
        task_description = ""
        try:
            if "llm_description" in f.attrs:
                if "which_llm_description" in f.attrs:
                    which_desc = f.attrs["which_llm_description"]
                    if int(which_desc) == 2 and "llm_description2" in f.attrs:
                        desc = f.attrs["llm_description2"]
                    else:
                        desc = f.attrs["llm_description"]
                else:
                    desc = f.attrs["llm_description"]
                if isinstance(desc, bytes):
                    task_description = desc.decode("utf-8")
                else:
                    task_description = str(desc)
        except Exception as e:
            print(f"Warning: Could not load task description from {hdf5_path}: {e}")
        pose_data = _extract_pose_actions(f)
    return pose_data, task_description


def _extract_pose_actions(hdf5_file) -> np.ndarray:
    """Extract pose actions (positions) from EgoDex HDF5."""
    actions: list[np.ndarray] = []
    pose_keys = [
        "transforms/leftHand",
        "transforms/rightHand",
        "transforms/leftIndexFingerTip",
        "transforms/rightIndexFingerTip",
        "transforms/camera",
    ]
    for key in pose_keys:
        if key in hdf5_file:
            transform_data = hdf5_file[key][:]  # (N, 4, 4)
            positions = transform_data[:, :3, 3]
            actions.append(positions)
    if not actions:
        if "transforms/camera" in hdf5_file:
            camera_transforms = hdf5_file["transforms/camera"][:]
            camera_positions = camera_transforms[:, :3, 3]
            actions.append(camera_positions)
        else:
            print("Warning: No pose data found, creating dummy actions")
            actions.append(np.zeros((100, 3)))
    return np.concatenate(actions, axis=1)


def load_egodex_dataset(dataset_path: str, max_trajectories: int = 100) -> dict[str, list[dict]]:
    """Load EgoDex dataset and organize by task, without a separate iterator class."""
    print(f"Loading EgoDex dataset from: {dataset_path}")
    print("=" * 100)
    print("LOADING EGODEX DATASET")
    print("=" * 100)

    dataset_path = Path(os.path.expanduser(dataset_path))
    if not dataset_path.exists():
        raise FileNotFoundError(f"Dataset path not found: {dataset_path}")

    traj_files = _discover_trajectory_files(dataset_path)
    print(f"Found {len(traj_files)} trajectory pairs")

    task_data: dict[str, list[dict]] = {}
    loaded_count = 0

    for hdf5_path, mp4_path, task_name in tqdm(traj_files, desc="Processing trajectories"):
        if max_trajectories is not None and loaded_count >= max_trajectories and max_trajectories != -1:
            break
        pose_data, task_description = _load_hdf5_data(hdf5_path)

        if "description unavailable" in task_description.lower():
            print(f"Skipping task {hdf5_path} because description is: {task_description}")
            continue
        frame_loader = EgoDexFrameLoader(str(mp4_path))

        assert task_description is not None

        trajectory = {
            "frames": frame_loader,
            # "actions": pose_data,
            "is_robot": False,
            "task": task_description,
            "quality_label": "successful",
            "preference_group_id": None,
            "preference_rank": None,
            "task_name": task_name,
            "id": generate_unique_id(),
        }

        task_data.setdefault(task_name, []).append(trajectory)
        loaded_count += 1

    total_trajectories = sum(len(v) for v in task_data.values())
    print(f"Loaded {total_trajectories} trajectories from {len(task_data)} tasks")

    return task_data