File size: 7,558 Bytes
7f97480 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 | {
"cells": [
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## 1. Load and read the dataset"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Read the dataset from hdf5 file and show the data struction"
]
},
{
"cell_type": "code",
"execution_count": 2,
"metadata": {},
"outputs": [],
"source": [
"import h5py\n",
"import json\n",
"import os\n",
"import random\n",
"\n",
"def get_all_hdf5_files(directory):\n",
" \"\"\"\n",
" Get all HDF5 files in a directory and its subdirectories.\n",
" \"\"\"\n",
" hdf5_files = []\n",
" for root, dirs, files in os.walk(directory):\n",
" for file in files:\n",
" if file.endswith('.hdf5'):\n",
" hdf5_files.append(os.path.join(root, file))\n",
" return hdf5_files\n",
"\n",
"def print_structure(group, indent=0):\n",
" \"\"\"\n",
" Print the hierarchical structure of an HDF5 group.\n",
" \"\"\"\n",
" for key in group.keys():\n",
" item = group[key]\n",
" print(\" \" * indent + key)\n",
" if isinstance(item, h5py.Group):\n",
" print_structure(item, indent + 4)\n",
" elif isinstance(item, h5py.Dataset):\n",
" print(\" \" * indent, item.shape, item.dtype)\n",
" if key == \"entities\" or key == \"target_entity\" or key == \"instruction\":\n",
" decoded = [x.decode('utf-8') for x in item]\n",
" print(\" \" * indent, decoded)\n",
" \n",
" elif isinstance(item, h5py.Datatype):\n",
" print(\" \" * (indent + 4) + \"Datatype\", json.loads())"
]
},
{
"cell_type": "code",
"execution_count": 2,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"['/media/shiduo/LENOVO_USB_HDD/dataset/VLABench/select_billiards/data_0.hdf5']\n"
]
}
],
"source": [
"dataset_root = \"/media/shiduo/LENOVO_USB_HDD/dataset/VLABench/select_billiards\" \n",
"hdf5_files = get_all_hdf5_files(dataset_root)\n",
"print(hdf5_files[:3])"
]
},
{
"cell_type": "code",
"execution_count": 4,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"data\n",
" 2025-02-23 20:46:40\n",
" instruction\n",
" (1,) |S38\n",
" ['Please put the striped_10 in any hole.']\n",
" meta_info\n",
" entities\n",
" (6,) |S15\n",
" ['billiards_table', 'striped_10', 'striped_14', 'striped_11', 'striped_12', 'solid_1']\n",
" target_entity\n",
" (1,) |S10\n",
" ['striped_10']\n",
" observation\n",
" depth\n",
" (212, 4, 480, 480) float32\n",
" ee_state\n",
" (212, 8) float32\n",
" point_cloud_colors\n",
" (212, 11905, 3) float32\n",
" point_cloud_points\n",
" (212, 11905, 3) float32\n",
" q_acceleration\n",
" (212, 7, 1) float32\n",
" q_state\n",
" (212, 7, 1) float32\n",
" q_velocity\n",
" (212, 7, 1) float32\n",
" rgb\n",
" (212, 4, 480, 480, 3) uint8\n",
" robot_mask\n",
" (212, 4, 480, 480) float32\n",
" trajectory\n",
" (212, 8) float32\n",
" 2025-02-23 20:49:53\n",
" instruction\n",
" (1,) |S38\n",
" ['Please put the striped_14 in any hole.']\n",
" meta_info\n",
" entities\n",
" (5,) |S15\n",
" ['billiards_table', 'striped_14', 'solid_7', 'striped_15', 'solid_1']\n",
" episode_config\n",
" () |S1879\n",
" target_entity\n",
" (1,) |S10\n",
" ['striped_14']\n",
" observation\n",
" depth\n",
" (164, 4, 480, 480) float32\n",
" ee_state\n",
" (164, 8) float32\n",
" point_cloud_colors\n",
" (164, 11929, 3) float32\n",
" point_cloud_points\n",
" (164, 11929, 3) float32\n",
" q_acceleration\n",
" (164, 7, 1) float32\n",
" q_state\n",
" (164, 7, 1) float32\n",
" q_velocity\n",
" (164, 7, 1) float32\n",
" rgb\n",
" (164, 4, 480, 480, 3) uint8\n",
" robot_mask\n",
" (164, 4, 480, 480) float32\n",
" trajectory\n",
" (164, 8) float32\n"
]
}
],
"source": [
"example_file = random.choice(hdf5_files)\n",
"with h5py.File(example_file, 'r') as f:\n",
" print_structure(f)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## 2. Convert to tf dataset"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"In the baseline methods, OpenVLA and Octo are trained using datasets in the TFDS format. In this repository, we have modified the tfds dataset builder to convert the HDF5 source format of VLABench into TFDS format. We offer [singlethread tfds builder](../VLABench/utils/rlds_builder.py) and [multithread tfds builder](../VLABench/utils/multithread_rlds_builder.py)."
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"To convert the dataset, run\n",
"```sh\n",
"python scripts/convert_to_rlds.py --save_dir /your/vlabench/dataset/root --task {target_task}\n",
"```\n",
"to create the tfds builder in the dataset directory.\n",
"Then, run\n",
"```sh\n",
"cd /your/vlabench/dataset/root/target_task\n",
"tfds build --overwrite\n",
"```\n",
"to generate rlds format dataset."
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"We recommend the readers to refer to https://github.com/kpertsch/rlds_dataset_builder for further guidance."
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## 3. Convert to Lerobot dataset"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"In the baseline methods, $\\pi_0$ uses lerobot format dataset for finetining. We offer a script to convert hdf5 format dataset into lerobot format, similar to Libero dataset used in $\\pi_0$.\n",
"\n",
"Run \n",
"```sh\n",
"python scripts/convert_to_lerobot.py --dataset-name xxx --dataset-path /your/path/to/hdf5 --max-files 500 --task-list task1 task2 ... task n\n",
"```\n",
"to create a multi-task lerobot dataset."
]
}
],
"metadata": {
"kernelspec": {
"display_name": "vlabench",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.10.4"
}
},
"nbformat": 4,
"nbformat_minor": 2
}
|