Add files using upload-large-folder tool

Surgical/utenn/benchtop_datasets_round2_with_part_seg/README.md

@@ -0,0 +1,227 @@
+<!--
+Open-H Embodiment Dataset README Template (v1.0)
+Please fill out this template and include it in the ./metadata directory of your LeRobot dataset.
+This file helps others understand the context and details of your contribution.
+-->
+
+# Benchtop Surgical Tool Manipulation Dataset (Open-H) - README
+
+---
+
+## 📋 At a Glance
+
+Bench-top surgical-tool video episodes with aligned **RGB**, **per-frame depth** (NPZ + visualization video), **part segmentation label maps**, **2D tool poses**, and optional **dynamic 3D Gaussian** scene/tool representations.
+
+---
+
+## 📖 Dataset Overview
+
+This dataset packages labeled benchtop clips into a **LeRobot dataset v2.1-style** directory layout that is **submit-ready for Open-H**. Each episode corresponds to a single recorded clip under an action label (task). For every timestep, we provide consistent indexing and references to raw depth/segmentation assets, plus per-tool 2D pose states.
+
+| | |
+| :--- | :--- |
+| **Total Trajectories** | `2` |
+| **Total Hours** | `Around 25 seconds` |
+| **Data Type** | `[ ] Clinical` `[ ] Ex-Vivo` `[x] Table-Top Phantom` `[ ] Digital Simulation` `[ ] Physical Simulation` `[ ] Other (If checked, update "Other")` |
+| **License** | CC BY 4.0 |
+| **Version** | `1.0` |
+
+---
+
+## 🧭 Dataset Structure & Navigation
+
+This dataset follows the LeRobot v2.1-style layout (manual writer) with **videos**, **per-episode parquet tables**, and **raw assets**.
+
+```
+<dataset_root>/
+  data/
+    chunk-000/
+      episode_000000.parquet
+      episode_000001.parquet
+      ...
+  videos/
+    chunk-000/
+      observation.images.rgb/        episode_000000.mp4 ...
+      observation.images.depth_vis/  episode_000000.mp4 ...
+      observation.images.part_id/    episode_000000.mp4 ...
+  meta/
+    info.json
+    tasks.jsonl
+    episodes.jsonl
+    episodes_stats.jsonl
+  metadata/
+    README.md   <-- this file
+  raw/
+    depth/episode_000000/frame_000000.npz ...
+    seg/episode_000000/00000.png ...
+    toolposes/episode_000000/tool_poses.json
+    toolposes/episode_000000/tool_name_to_slot.json
+    seg_schema/episode_000000/label_schema.json
+    seg_schema/episode_000000/label_id_maps.json
+    gaussians/episode_000000/dynamic_gaussians_frames.jsonl.gz   (optional)
+    gaussians/episode_000000/dynamic_gaussians_meta.json         (optional)
+    gaussians/episode_000000/photometric_cam_T_wc.json           (optional)
+```
+
+### Where to look first
+- **meta/info.json**: overall dataset metadata (fps, splits, modalities, counts)
+- **meta/tasks.jsonl**: list of action labels / tasks
+- **meta/episodes.jsonl**: episode index → clip id, paths, tool list, gaussians presence
+- **data/chunk-000/episode_XXXXXX.parquet**: per-timestep table (timestamps, tool poses, raw pointers)
+- **videos/**: fast visualization for RGB / depth_vis / part_id
+- **raw/**: submission-grade assets (relative depth NPZ, label-map PNGs, toolposes JSON, optional gaussians)
+
+---
+
+## 🎯 Tasks & Domain
+
+### Domain
+- [x] **Surgical Robotics**
+- [ ] **Ultrasound Robotics**
+- [ ] **Other Healthcare Robotics** (Please specify: `N/A`)
+
+### Demonstrated Skills
+Action labels are stored in:
+- `meta/tasks.jsonl` (task definitions)
+- `action.label` column in each parquet
+
+Examples (replace with your actual labels):
+- Tool manipulation (grasp / pick-and-place)
+- Retraction / interaction
+- Cutting / dissection-like motions
+
+---
+
+## 🔬 Data Collection Details
+
+### Collection Method
+- [x] **Human Teleoperation** *(typical for benchtop surgical tool demos; update if different)*
+- [ ] **Programmatic/State-Machine**
+- [ ] **AI Policy / Autonomous**
+- [ ] **Other** (Please specify: `TBD`)
+
+### Operator Details
+
+| | Description |
+| :--- | :--- |
+| **Operator Count** | `N/A` |
+| **Operator Skill Level** | `[ ] Expert (e.g., Surgeon, Sonographer)` <br> `[x] Intermediate (e.g., Trained Researcher)` <br> `[ ] Novice (e.g., ML Researcher with minimal experience)` <br> `[ ] N/A` |
+| **Collection Period** | `N/A` |
+
+### Recovery Demonstrations
+- [ ] **Yes**
+- [x] **No**
+
+If yes, describe: `TBD`
+
+---
+
+## 💡 Diversity Dimensions
+
+- [ ] **Camera Position / Angle**
+- [ ] **Lighting Conditions**
+- [ ] **Target Object**
+- [ ] **Spatial Layout**
+- [ ] **Robot Embodiment**
+- [ ] **Task Execution**
+- [ ] **Background / Scene**
+- [ ] **Other** (Please specify: `TBD`)
+
+Elaboration: `TBD`
+
+---
+
+## 🛠️ Equipment & Setup
+
+
+### Sensors & Cameras
+
+| Type | Model/Details |
+| :--- | :--- |
+| **Primary Camera** | `Monocular RGB video (resolution in meta/episodes_stats.jsonl)` |
+| **Depth** | `Per-frame depth stored as NPZ (raw/depth) + depth_vis mp4 (videos/...)` |
+| **Segmentation** | `Part/label-map PNGs (raw/seg) + part_id mp4 (videos/...)` |
+| **Other** | `Dynamic 3D Gaussians (optional) in raw/gaussians/` |
+
+---
+
+## 🎯 Action & State Space Representation
+
+This dataset is primarily **vision + tool-state** oriented. Robot kinematics may not be included.
+
+### Action Space Representation
+
+**Primary Action Representation:**
+- [ ] **Absolute Cartesian**
+- [ ] **Relative Cartesian**
+- [ ] **Joint Space**
+- [x] **Other**: *Action is a categorical label per episode (`action.label`)*
+
+**Action Dimensions:**
+```
+action.label: string
+- Categorical action label (same for all timesteps in an episode)
+```
+
+### State Space Representation
+
+**State Information Included:**
+- [ ] **Joint Positions**
+- [ ] **Joint Velocities**
+- [ ] **End-Effector Pose**
+- [ ] **Force/Torque Readings**
+- [ ] **Gripper State**
+- [x] **Other**: *2D tool pose state (per tool)*
+
+**State Dimensions:**
+```
+observation.state.tool_pose_2d: [num_tools, 3]
+- tx_px: tool center / tip x position in pixels (see toolposes JSON definition)
+- ty_px: tool center / tip y position in pixels
+- theta_rad: in-plane rotation (radians)
+```
+
+Additional related fields in parquet:
+- `observation.meta.tool_name`: list of tool names (ordering reference)
+- `observation.meta.tool_visible`: per-tool boolean mask
+- `observation.meta.depth_npz_relpath`: pointer to raw depth NPZ per timestep
+- `observation.meta.seg_png_relpath`: pointer to raw part label-map PNG per timestep
+
+### Dynamic 3D Gaussians (Optional)
+
+If present, per-episode files live under:
+- `raw/gaussians/episode_XXXXXX/dynamic_gaussians_frames.jsonl.gz`
+- `raw/gaussians/episode_XXXXXX/dynamic_gaussians_meta.json`
+- `raw/gaussians/episode_XXXXXX/photometric_cam_T_wc.json`
+
+These encode per-frame 3D Gaussian parameters (positions/covariances/colors/opacities, etc. as reconstructed by MoSca https://github.com/JiahuiLei/MoSca) as benchtop scene representations and camera transforms (intrinsics and extrinsics) for 3D reconstruction or tracking research.
+
+---
+
+## ⏱️ Data Synchronization Approach
+
+All modalities are aligned **by frame index** during export.
+
+- The RGB video provides the reference frame count.
+- Depth NPZ frames, segmentation PNG frames, and tool pose frames are truncated to:
+  ```
+  n_steps = min(rgb_frames, depth_npz_count, seg_png_count, toolposes_num_frames)
+  ```
+- Timestamps are generated as:
+  ```
+  timestamp_s = step_index / fps_timestamps
+  ```
+  where `fps_timestamps` is recorded in `meta/info.json`.
+
+This approach ensures consistent timestep alignment across all exported modalities.
+
+---
+
+## 👥 Attribution & Contact
+
+| | |
+| :--- | :--- |
+| **Dataset Lead** | Nan Xiao, Tom Olesch, Farong Wang, Fei Liu, DeLong Jonathan C. |
+| **Institution** | University of Tennessee, Knoxville (AURAS Lab) |
+| **Contact Email** | nxiao4@vols.utk.edu |
+| **Citation (BibTeX)** | <pre><code>@misc{benchtop_openh_2026,<br>  author = {Xiao, Nan and collaborators},<br>  title = {Benchtop Surgical Tool Manipulation Dataset (Open-H)},<br>  year = {2026},<br>  publisher = {Open-H-Embodiment},<br>  note = {Dataset packaging in LeRobot v2.1-style layout}<br>}</code></pre> |

Surgical/utenn/benchtop_datasets_round2_with_part_seg/meta/episodes.jsonl

@@ -0,0 +1,2 @@
+{"episode_index": 0, "episode_id": "episode_000000", "task": "PULLING", "clip_id": "t2_video", "fps_timestamps": 10.0, "fps_videos": 10.0, "num_steps": 127, "duration_s": 12.6, "videos": {"observation.images.rgb": "videos/chunk-000/observation.images.rgb/episode_000000.mp4", "observation.images.depth_vis": "videos/chunk-000/observation.images.depth_vis/episode_000000.mp4", "observation.images.part_id": "videos/chunk-000/observation.images.part_id/episode_000000.mp4"}, "parquet": "data/chunk-000/episode_000000.parquet", "raw_paths": {"raw_toolposes_dir": "raw/toolposes/episode_000000", "raw_depth_dir": "raw/depth/episode_000000", "raw_seg_dir": "raw/seg/episode_000000", "raw_seg_schema_dir": "raw/seg_schema/episode_000000", "raw_gaussians_dir": "raw/gaussians/episode_000000"}, "tools": {"tool_names": ["grasper", "grasper2"], "tool_name_to_slot": {"grasper": 0, "grasper2": 1}, "pose_definition": "[tx_px, ty_px, theta_rad] per tool; NaN if not visible"}, "modalities_present": {"depth": true, "segmentation": true, "dynamic_gaussians": true}, "dynamic_gaussians": {"present": true, "frames_jsonl_gz": "raw/gaussians/episode_000000/dynamic_gaussians_frames.jsonl.gz", "meta_json": "raw/gaussians/episode_000000/dynamic_gaussians_meta.json", "cam_T_wc_json": "raw/gaussians/episode_000000/photometric_cam_T_wc.json"}}
+{"episode_index": 1, "episode_id": "episode_000001", "task": "PULLING", "clip_id": "t4_video", "fps_timestamps": 10.0, "fps_videos": 10.0, "num_steps": 96, "duration_s": 9.5, "videos": {"observation.images.rgb": "videos/chunk-000/observation.images.rgb/episode_000001.mp4", "observation.images.depth_vis": "videos/chunk-000/observation.images.depth_vis/episode_000001.mp4", "observation.images.part_id": "videos/chunk-000/observation.images.part_id/episode_000001.mp4"}, "parquet": "data/chunk-000/episode_000001.parquet", "raw_paths": {"raw_toolposes_dir": "raw/toolposes/episode_000001", "raw_depth_dir": "raw/depth/episode_000001", "raw_seg_dir": "raw/seg/episode_000001", "raw_seg_schema_dir": "raw/seg_schema/episode_000001", "raw_gaussians_dir": "raw/gaussians/episode_000001"}, "tools": {"tool_names": ["grasper", "grasper2"], "tool_name_to_slot": {"grasper": 0, "grasper2": 1}, "pose_definition": "[tx_px, ty_px, theta_rad] per tool; NaN if not visible"}, "modalities_present": {"depth": true, "segmentation": true, "dynamic_gaussians": true}, "dynamic_gaussians": {"present": true, "frames_jsonl_gz": "raw/gaussians/episode_000001/dynamic_gaussians_frames.jsonl.gz", "meta_json": "raw/gaussians/episode_000001/dynamic_gaussians_meta.json", "cam_T_wc_json": "raw/gaussians/episode_000001/photometric_cam_T_wc.json"}}

Surgical/utenn/benchtop_datasets_round2_with_part_seg/meta/episodes_stats.jsonl

@@ -0,0 +1,2 @@
+{"episode_index": 0, "action": "PULLING", "clip_id": "t2_video", "num_steps": 127, "duration_s": 12.6, "fps_timestamps": 10.0, "fps_videos": 10.0, "rgb_hw": [416, 720], "num_tools": 2, "tool_names": ["grasper", "grasper2"], "depth_present": true, "seg_present": true, "gaussians_present": true}
+{"episode_index": 1, "action": "PULLING", "clip_id": "t4_video", "num_steps": 96, "duration_s": 9.5, "fps_timestamps": 10.0, "fps_videos": 10.0, "rgb_hw": [416, 720], "num_tools": 2, "tool_names": ["grasper", "grasper2"], "depth_present": true, "seg_present": true, "gaussians_present": true}

Surgical/utenn/benchtop_datasets_round2_with_part_seg/meta/info.json

@@ -0,0 +1,75 @@
+{
+  "repo_id": "utk/benchtop_v1",
+  "dataset_format_version": "2.1",
+  "lerobot_version_target": "0.3.3",
+  "license": "CC BY 4.0",
+  "robot_type": "unknown",
+  "use_videos": true,
+  "fps_timestamps": 10.0,
+  "fps_videos": 10.0,
+  "modalities": [
+    "observation.images.rgb",
+    "observation.images.depth_vis",
+    "observation.images.part_id"
+  ],
+  "additional_fields": {
+    "observation.meta.tool_name": {
+      "dtype": "string",
+      "shape": [
+        "num_tools"
+      ],
+      "description": "Tool name list; defines ordering for tool-visible and 2D pose arrays."
+    },
+    "observation.meta.tool_visible": {
+      "dtype": "bool",
+      "shape": [
+        "num_tools"
+      ],
+      "description": "Per-tool visibility flag for the 2D pose."
+    },
+    "observation.state.tool_pose_2d": {
+      "dtype": "float32",
+      "shape": [
+        "num_tools",
+        3
+      ],
+      "names": [
+        "tx_px",
+        "ty_px",
+        "theta_rad"
+      ],
+      "description": "Per-tool 2D pose [tx,ty,theta] aligned with tool_name ordering. NaN if not visible."
+    },
+    "observation.meta.depth_npz_relpath": {
+      "dtype": "string",
+      "shape": [
+        1
+      ],
+      "description": "Relative path to raw depth .npz under raw/depth/episode_xxxxxx/ (only present when depth exists)."
+    },
+    "observation.meta.seg_png_relpath": {
+      "dtype": "string",
+      "shape": [
+        1
+      ],
+      "description": "Relative path to raw label_map png under raw/seg/episode_xxxxxx/ (only present when segmentation exists)."
+    }
+  },
+  "splits": {
+    "train": "0:1",
+    "val": "1:1",
+    "test": "1:2"
+  },
+  "num_episodes": 2,
+  "chunking": {
+    "chunks": [
+      "chunk-000"
+    ]
+  },
+  "notes": {
+    "truncation_policy": "n_steps = min(counts over available modalities: rgb, toolposes, and optionally depth/seg if present)",
+    "depth_storage": "raw npz per frame + depth_vis video for viewing (only if depth is available)",
+    "seg_storage": "raw label_map png per frame + part_id video for viewing (only if segmentation is available)",
+    "gaussians_storage": "episode-level raw copy if present"
+  }
+}

Surgical/utenn/benchtop_datasets_round2_with_part_seg/meta/tasks.jsonl

@@ -0,0 +1 @@
+{"task": "PULLING", "description": "Benchtop action label: PULLING"}