Upload folder using huggingface_hub

rc_datasets/arx5_joint/convert_arx5_joint_to_lerobot_v21.py

@@ -0,0 +1,257 @@
+"""
+Minimal example: convert dataset to the LeRobot format.
+
+CLI Example (using the *arrange_flowers* task as an example):
+    python convert_libero_to_lerobot.py \
+        --repo-name arrange_flowers_repo \
+        --raw-dataset /path/to/arrange_flowers \
+        --frame-interval 1 \
+
+Notes:
+- If you plan to push to the Hugging Face Hub later, handle that outside this script.
+"""
+
+import argparse
+import json
+import shutil
+from pathlib import Path
+from typing import Any, Dict, List
+
+import cv2
+import numpy as np
+from lerobot.datasets.lerobot_dataset import HF_LEROBOT_HOME, LeRobotDataset
+
+def load_jsonl(path: Path) -> List[Dict[str, Any]]:
+    """Load a JSONL file into a list of dicts."""
+    with path.open("r", encoding="utf-8") as f:
+        return [json.loads(line) for line in f]
+
+
+def create_lerobot_dataset(
+    repo_name: str,
+    robot_type: str,
+    fps: float,
+    height: int,
+    width: int,
+) -> LeRobotDataset:
+    """
+    Create a LeRobot dataset with custom feature schema
+    """
+    dataset = LeRobotDataset.create(
+        repo_id=repo_name,
+        robot_type=robot_type,
+        fps=fps,
+        features={
+            "global_image": {
+                "dtype": "video",
+                "shape": (height, width, 3),
+                "names": ["height", "width", "channel"],
+            },
+            "wrist_image": {
+                "dtype": "video",
+                "shape": (height, width, 3),
+                "names": ["height", "width", "channel"],
+            },
+            "right_image": {
+                "dtype": "video",
+                "shape": (height, width, 3),
+                "names": ["height", "width", "channel"],
+            },
+            "state": {
+                "dtype": "float32",
+                "shape": (7,), # for ee_pose and gripper width
+                "names": ["state"],
+            },
+            "actions": {
+                "dtype": "float32",
+                "shape": (7,), # for ee_pose and gripper width
+                "names": ["actions"],
+            },
+        },
+        image_writer_threads=32,
+        image_writer_processes=16,
+    )
+    return dataset
+
+
+def process_episode_dir(
+    episode_path: Path,
+    dataset: LeRobotDataset,
+    frame_interval: int,
+    prompt: str,
+) -> None:
+    """
+    Process a single episode directory and append frames to the given dataset.
+
+    episode_path : Path
+        Episode directory containing `states/states.jsonl` and `videos/*.mp4`.
+    dataset : LeRobotDataset
+        Target dataset to which frames are added.
+    frame_interval : int
+        Sampling stride (>=1).
+    prompt : str
+        Language instruction of this episode.
+    """
+    # Modify if your dataset consists of bimanual data.
+    states_path = episode_path / "states" / "states.jsonl" 
+    videos_dir = episode_path / "videos"
+
+    ep_states = load_jsonl(states_path)
+
+    # adjust them to match your dataset’s actual naming.
+    wrist_video = cv2.VideoCapture(str(videos_dir / "arm_realsense_rgb.mp4")) 
+    global_video = cv2.VideoCapture(str(videos_dir / "global_realsense_rgb.mp4"))
+    right_video  = cv2.VideoCapture(str(videos_dir / "right_realsense_rgb.mp4"))
+
+    wrist_frames_count  = int(wrist_video.get(cv2.CAP_PROP_FRAME_COUNT))
+    global_frames_count = int(global_video.get(cv2.CAP_PROP_FRAME_COUNT))
+    right_frames_count  = int(right_video.get(cv2.CAP_PROP_FRAME_COUNT))
+    n_states = len(ep_states)
+
+    # assert all lengths match 
+    assert (
+        n_states == wrist_frames_count == global_frames_count == right_frames_count
+    ), (
+        f"Mismatch in episode {episode_path.name}: "
+        f"states={n_states}, wrist={wrist_frames_count}, "
+        f"global={global_frames_count}, right={right_frames_count}"
+    )
+
+    # write frames to the episode of lerobot dataset
+    for idx in range(frame_interval, n_states, frame_interval):
+        # Build pose
+        pose = np.concatenate(
+            (np.asarray(ep_states[idx]["joint_positions"]), [ep_states[idx]["gripper_width"]])
+        )
+        last_pose = np.concatenate(
+            (np.asarray(ep_states[idx - frame_interval]["joint_positions"]),
+             [ep_states[idx - frame_interval]["gripper_width"]])
+        )
+
+        # Read frames && BGR -> RGB
+        # Resize as needed, but update the LeRobot feature shape accordingly.
+        _, wrist_image  = wrist_video.read()
+        _, global_image = global_video.read()
+        _, right_image  = right_video.read()
+        
+        wrist_image  = cv2.cvtColor(wrist_image,  cv2.COLOR_BGR2RGB)
+        global_image = cv2.cvtColor(global_image, cv2.COLOR_BGR2RGB)
+        right_image  = cv2.cvtColor(right_image,  cv2.COLOR_BGR2RGB)
+
+        dataset.add_frame(
+            {
+                "global_image": global_image,
+                "wrist_image": wrist_image,
+                "right_image": right_image,
+                "state": last_pose.astype(np.float32, copy=False),
+                "actions": pose.astype(np.float32, copy=False),
+            },
+            task=prompt
+        )
+
+    wrist_video.release()
+    global_video.release()
+    right_video.release()
+    dataset.save_episode()
+
+
+def main(
+    repo_name: str,
+    raw_dataset: Path,
+    frame_interval: int = 1,
+    overwrite_repo: bool = False,
+) -> None:
+    """
+    Convert a dataset directory into LeRobot format.
+
+    repo_name : str
+        Output repo/dataset name (saved under $HF_LEROBOT_HOME / repo_name).
+    raw_dataset : Path
+        Path to the raw dataset root directory.
+    frame_interval : int, default=1
+        Sample every N frames (kept identical).
+    overwrite_repo : bool, default=False
+        If True, remove the existing dataset directory before writing.
+    """
+    assert frame_interval >= 1, "frame_interval must be >= 1"
+
+    # overwrite repo
+    dst_dir = HF_LEROBOT_HOME / repo_name
+    if overwrite_repo and dst_dir.exists():
+        print(f"removing existing dataset at {dst_dir}")
+        shutil.rmtree(dst_dir)
+
+    # Load task_infos
+    task_info_path = raw_dataset / "meta" / "task_info.json"
+    with task_info_path.open("r", encoding="utf-8") as f:
+        task_info = json.load(f)
+
+    robot_type = task_info["task_desc"]["task_tag"][2]  # "ARX5"
+    video_info = task_info["video_info"]
+    video_info["width"]  = 640  # TODO: derive from task_info or actual videos
+    video_info["height"] = 480
+    fps = video_info["fps"]
+
+    prompt = task_info["task_desc"]["prompt"]
+
+    # Create dataset, define feature in the form you need.
+    # - proprio is stored in `state` and actions in `action`
+    # - LeRobot assumes that dtype of image data is `image`
+    dataset = create_lerobot_dataset(
+        repo_name=repo_name,
+        robot_type=robot_type,
+        fps=fps,
+        height=video_info["height"],
+        width=video_info["width"],
+    )
+
+    # populate the dataset to lerobot dataset
+    data_root = raw_dataset / "data"
+    for episode_path in data_root.iterdir():
+        if not episode_path.is_dir():
+            continue
+        print(f"Processing episode: {episode_path.name}")
+        process_episode_dir(
+            episode_path=episode_path,
+            dataset=dataset,
+            frame_interval=frame_interval,
+            prompt=prompt,
+        )
+    
+    print(f"Done. Dataset saved to: {dst_dir}")
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(
+        description="Convert a custom dataset to LeRobot format."
+    )
+    parser.add_argument(
+        "--repo-name",
+        required=True,
+        help="Name of the output dataset (under $HF_LEROBOT_HOME).",
+    )
+    parser.add_argument(
+        "--raw-dataset",
+        required=True,
+        type=str,
+        help="Path to the raw dataset root.",
+    )
+    parser.add_argument(
+        "--frame-interval",
+        type=int,
+        default=1,
+        help="Sample every N frames. Default: 1",
+    )
+    parser.add_argument(
+        "--overwrite-repo",
+        action="store_true",
+        help="Remove existing output directory if it exists.",
+    )
+    args = parser.parse_args()
+    
+    main(
+        repo_name=args.repo_name,
+        raw_dataset=Path(args.raw_dataset),
+        frame_interval=args.frame_interval,
+        overwrite_repo=args.overwrite_repo,
+    )

rc_datasets/arx5_joint/rc_arx5_joint_arrange_flowers.tar.gz.aa

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9f40d43a0cf078483ef26184eabd394d43339fa353cb3c74a69250813df0bff5
+size 10737418240

rc_datasets/arx5_joint/rc_arx5_joint_arrange_flowers.tar.gz.ab

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1f6ad91b6b109a1cd4db60d202eb4617044faa234f2d82e5c861c41e285c74f9
+size 10737418240

rc_datasets/arx5_joint/rc_arx5_joint_arrange_flowers.tar.gz.ac

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bd05ea866b0ef5f6056cb1ede3cf1a92dc32150311651b2c830a7802634aa532
+size 10737418240

rc_datasets/arx5_joint/rc_arx5_joint_arrange_flowers.tar.gz.ad

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ad5ffec412ece2349b889ae19f40a1a5730943a0a69d805cbbf902fa7e34b579
+size 10737418240

rc_datasets/arx5_joint/rc_arx5_joint_arrange_flowers.tar.gz.ae

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:86db971c176987361e7b6d7f9c12e7217a379d6ff8e7f7c3681c16b72f4d57a5
+size 10737418240

rc_datasets/arx5_joint/rc_arx5_joint_arrange_flowers.tar.gz.af

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fb4fd8c63ddc3dc4d77a3c0570013bc787e101ba640abc1b45e65dab965fc880
+size 3381024569

rc_datasets/arx5_joint/rc_arx5_joint_arrange_paper_cups.tar.gz.aa

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d82e8310ba38bbca0d930be6ddd1bfb37c15e2ff5514a1def795326da34f49ed
+size 10737418240

rc_datasets/arx5_joint/rc_arx5_joint_arrange_paper_cups.tar.gz.ab

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2bd2ee8c1874fd2c89f25cfa5b9dddda010f7ff1a0f1d0d692da793815e25d08
+size 10737418240

rc_datasets/arx5_joint/rc_arx5_joint_arrange_paper_cups.tar.gz.ac

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:759b6779fb540bb0b797a302d102e9eb9328a796bd14689e91feb2582f098e47
+size 10737418240

rc_datasets/arx5_joint/rc_arx5_joint_arrange_paper_cups.tar.gz.ad

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:663b1ce2e041c9acf4b9e23c2682cd12c668bad8fc73628d75048a880ce08779
+size 2559421500

rc_datasets/arx5_joint/rc_arx5_joint_fold_dishcloth.tar.gz.aa

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c36a58e99f8c5d8cc02c8570dc840efd4b10d323c8940cf0e10d151c80c56386
+size 10737418240

rc_datasets/arx5_joint/rc_arx5_joint_fold_dishcloth.tar.gz.ab

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e284c5b3bdc5b8fe6c4215b3b80039252c86545025d8b03c79cf32db28c14c65
+size 10737418240

rc_datasets/arx5_joint/rc_arx5_joint_fold_dishcloth.tar.gz.ac

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6229691b3fcf638ffd9bdf3196c406d211fec5f64f0ddc3dc39c0c8447f221b9
+size 7505854771

rc_datasets/arx5_joint/rc_arx5_joint_open_the_drawer.tar.gz.aa

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bf55118089f0c79f1ddd2d34e35fd7ac336a35409bc509382e0a2e7a16fe546a
+size 9241840307

rc_datasets/arx5_joint/rc_arx5_joint_place_shoes_on_rack.tar.gz.aa

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e5a6eaa6ae9b903b9f20db680b68eb047ddc9b496a6fae5bf68eaf169018bf7f
+size 10737418240

rc_datasets/arx5_joint/rc_arx5_joint_place_shoes_on_rack.tar.gz.ab

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0805274e29458319fd7f6f04a2bd68f96294d0196c344b4329b6e90327ef477e
+size 10737418240

rc_datasets/arx5_joint/rc_arx5_joint_place_shoes_on_rack.tar.gz.ac

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8e8e671284e663bda32b1187e615ecfa74e6354e33e9660a96468b6bec26eac3
+size 1190131940

rc_datasets/arx5_joint/rc_arx5_joint_put_cup_on_coaster.tar.gz.aa

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fe721c0cdca015c64080e80298d1d3d55b0a90784870d1dfb8cdda9d6a800e10
+size 6725006246

rc_datasets/arx5_joint/rc_arx5_joint_search_green_boxes.tar.gz.aa

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:89ac697a8a04dca2ef19e44e0721f1525d2058205eb7a8a4b436811294a02139
+size 10737418240

rc_datasets/arx5_joint/rc_arx5_joint_search_green_boxes.tar.gz.ab

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:02ebfc531c9cfec6591b2119a0623606f4faaf887a0d2000e8f0cc9ccc9ca6dd
+size 9619710191

rc_datasets/arx5_joint/rc_arx5_joint_sort_electronic_products.tar.gz.aa

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6c4f0dac22a4263ae468705b54226cdbadd19e298ce7bfa683b5c3d38bf85aad
+size 10737418240

rc_datasets/arx5_joint/rc_arx5_joint_sort_electronic_products.tar.gz.ab

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b26978406d2c2ae72b15c70eb5086bfe95dbe30185398e75f9856ae1c566828f
+size 10737418240

rc_datasets/arx5_joint/rc_arx5_joint_sort_electronic_products.tar.gz.ac

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:51e37188814508c6c2ee681231dc1eeaf0a833c880b27044fcd85572f500c94e
+size 10737418240

rc_datasets/arx5_joint/rc_arx5_joint_sort_electronic_products.tar.gz.ad

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:599051a8cdcce57358bd2997d744e8efb3143d46c5b82bedc86466a92af2ed72
+size 9037922761

rc_datasets/arx5_joint/rc_arx5_joint_turn_on_light_switch.tar.gz.aa

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1609b17aade5c106cd0c757607dcb14352f9cae71f8f707bf8c00ba83cf707af
+size 5541354983

rc_datasets/arx5_joint/rc_arx5_joint_water_potted_plant.tar.gz.aa

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7eb7ec133d747b1d773d8c9c7f7149bfd305ef9da9236b172a9d18279414ab81
+size 10737418240

rc_datasets/arx5_joint/rc_arx5_joint_water_potted_plant.tar.gz.ab

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4b32ded89f99c644c6853e84cfd60ee3a3b377f28c26ba5cacf2ae9e9976ca81
+size 10737418240

rc_datasets/arx5_joint/rc_arx5_joint_water_potted_plant.tar.gz.ac

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:27a097306050755120ecdd57d8345ca65fb5b84f0cce2d97b05d16810cc8c0dd
+size 3562037753

rc_datasets/arx5_joint/rc_arx5_joint_wipe_the_table.tar.gz.aa

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2d2db60aff684d8341b1254d9ff8c80f4d2ec4e6f20be4d55eb7e67a7089be4c
+size 10737418240

rc_datasets/arx5_joint/rc_arx5_joint_wipe_the_table.tar.gz.ab

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ab0af0239f0facf820824c2edff40bd22388e788acdf5eef052936e2d759a163
+size 10737418240

rc_datasets/arx5_joint/rc_arx5_joint_wipe_the_table.tar.gz.ac

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1b5cbe541d7e1c75c3067609c42d91245dd15a8d084b9bfe00d4d519962c78d9
+size 1448567610