Datasets:

nyuvideolab
/

danceNet3D

@@ -4,30 +4,3 @@
 *.cube filter=lfs diff=lfs merge=lfs -text
 *.png filter=lfs diff=lfs merge=lfs -text
 *.jpg filter=lfs diff=lfs merge=lfs -text
-3dgs/s4/3PointStep.zip filter=lfs diff=lfs merge=lfs -text
-3dgs/s4/BartSimpson.zip filter=lfs diff=lfs merge=lfs -text
-3dgs/s4/BizMarkie.zip filter=lfs diff=lfs merge=lfs -text
-3dgs/s4/HouseFootworkAdvanced.zip filter=lfs diff=lfs merge=lfs -text
-3dgs/s4/RoboCop.zip filter=lfs diff=lfs merge=lfs -text
-3dgs/s4/RunningMan.zip filter=lfs diff=lfs merge=lfs -text
-3dgs/s4/TheRooftop.zip filter=lfs diff=lfs merge=lfs -text
-3dgs/s4/ToeTaps.zip filter=lfs diff=lfs merge=lfs -text
-3dgs/s4/WuTang.zip filter=lfs diff=lfs merge=lfs -text
-3dgs/s5/AttitudePromenade.zip filter=lfs diff=lfs merge=lfs -text
-3dgs/s5/BasicSuzieQ.zip filter=lfs diff=lfs merge=lfs -text
-3dgs/s5/BigKicks.zip filter=lfs diff=lfs merge=lfs -text
-3dgs/s5/BourreeTurns2.zip filter=lfs diff=lfs merge=lfs -text
-3dgs/s5/Chacha.zip filter=lfs diff=lfs merge=lfs -text
-3dgs/s5/DoubleSpiral.zip filter=lfs diff=lfs merge=lfs -text
-3dgs/s5/Flair.zip filter=lfs diff=lfs merge=lfs -text
-3dgs/s5/Jumping.zip filter=lfs diff=lfs merge=lfs -text
-3dgs/s5/Pirouettes.zip filter=lfs diff=lfs merge=lfs -text
-3dgs/s5/Portdebras.zip filter=lfs diff=lfs merge=lfs -text
-3dgs/s5/RonDeJambeAtere2.zip filter=lfs diff=lfs merge=lfs -text
-3dgs/s5/SonBasic.zip filter=lfs diff=lfs merge=lfs -text
-3dgs/s6/BiancaGolden_Breathing.zip filter=lfs diff=lfs merge=lfs -text
-3dgs/s6/BiancaGolden_Chimee.zip filter=lfs diff=lfs merge=lfs -text
-3dgs/s6/BiancaGolden_CircleTurns.zip filter=lfs diff=lfs merge=lfs -text
-3dgs/s6/BiancaGolden_GrandPlies.zip filter=lfs diff=lfs merge=lfs -text
-3dgs/s6/BiancaGolden_SalsaBasic.zip filter=lfs diff=lfs merge=lfs -text
-3dgs/s6/RobertRubama_RussiaCostume.zip filter=lfs diff=lfs merge=lfs -text

 *.cube filter=lfs diff=lfs merge=lfs -text
 *.png filter=lfs diff=lfs merge=lfs -text
 *.jpg filter=lfs diff=lfs merge=lfs -text

README.md CHANGED Viewed

@@ -27,16 +27,16 @@ size_categories:
 # DanceNet3D
-A large-scale synchronized multi-view dance dataset captured with 28-29 calibrated cameras at 800x1280 resolution. The dataset contains 47 dance sequences across 3 recording sessions, totaling **35,509 frames** with per-frame camera calibration (COLMAP), foreground masks, and optional color correction LUTs.
 ## Dataset Summary
 | Session | Sequences (available / total) | Cameras | Frames (available / total) | Size |
 |---------|-------------------------------|---------|---------------------------|------|
-| s4 | 9 / 9 | 29 | 6,786 / 6,786 | ~44 GB |
-| s5 | 14 / 23 | 29 | 11,253 / 17,999 | ~53 GB |
-| s6 | 6 / 15 | 28 | 4,134 / 10,724 | ~19 GB |
-| **Total** | **29 / 47** | | **22,173 / 35,509** | **~116 GB** |
 ## Data Format
@@ -47,7 +47,8 @@ Each session is stored as a directory (`s4/`, `s5/`, `s6/`) containing per-seque
 - **Foreground masks**: `masks.tar.zst` binary person segmentation masks generated with SAM3 and with manual quality review
 - **Manifest**: `manifest.json` frame IDs, camera lists, and sequence metadata
-### Directory Structure
 ```
 DanceNet3D/
@@ -77,7 +78,10 @@ DanceNet3D/
     └── ...
 ```
-### Extracted Frame Structure
 After running `video_to_images.py`, the data is organized per-frame:
@@ -104,20 +108,18 @@ output/
         └── ...
 ```
 ## Quick Start
-### Prerequisites
-- Python 3.8+
-- FFmpeg
-- zstd
 ```bash
 pip install -r requirements.txt
 ```
-### Extract Frames
 ```bash
 # Extract a single session
 python video_to_images.py --input s5 --output extracted/s5
@@ -132,13 +134,34 @@ python video_to_images.py --input s5 --output extracted/s5 --cameras 0028 1362
 python video_to_images.py --input s5 --output extracted/s5 --apply-lut
 ```
 ### Using with COLMAP
 The `colmap/` directory in each sequence contains pre-computed camera intrinsics and extrinsics in COLMAP format. Camera parameters correspond to the undistorted, rotated (portrait orientation) images.
 ## Sequences
-### Session 4 (s4)
 | Sequence | Frames | Cameras | Status |
 |----------|--------|---------|--------|
@@ -152,7 +175,10 @@ The `colmap/` directory in each sequence contains pre-computed camera intrinsics
 | ToeTaps | 572 | 29 | Available |
 | WuTang | 884 | 29 | Available |
-### Session 5 (s5)
 | Sequence | Frames | Cameras | Status |
 |----------|--------|---------|--------|
@@ -180,7 +206,10 @@ The `colmap/` directory in each sequence contains pre-computed camera intrinsics
 | Turns | 733 | 29 | Coming soon |
 | Twists | 930 | 29 | Coming soon |
-### Session 6 (s6)
 | Sequence | Frames | Cameras | Status |
 |----------|--------|---------|--------|
@@ -200,24 +229,7 @@ The `colmap/` directory in each sequence contains pre-computed camera intrinsics
 | BiancaGolden_SyncopatedGroove | 919 | 28 | Coming soon |
 | RobertRubama_RussiaRehearsal | 727 | 28 | Coming soon |
-## How to View 3DGS
-Pre-computed 3D Gaussian Splatting (3DGS) reconstructions are available under the `/3dgs` directory. Each `.zip` file corresponds to one dance sequence and contains multiple `.ply` files — one per frame.
-**Download and decompress:**
-```bash
-unzip AttitudePromenade.zip
-```
-On Mac, you can also double-click the `.zip` file in Finder. On Windows, right-click and use a tool like 7-Zip or WinRAR.
-**View a single frame:**
-1. Go to [https://superspl.at/editor](https://superspl.at/editor)
-2. Upload a single `.ply` file
-**Playback multi-frame animation:**
-1. Go to [https://superspl.at/editor](https://superspl.at/editor)
-2. Upload all `.ply` files from the sequence at once
-3. Hit the **Play** button above the timeline
 ## Technical Details

 # DanceNet3D
+A large-scale synchronized multi-view dance dataset captured with 28-29 calibrated cameras at 800x1280 resolution. The dataset contains 47 dance sequences across 3 recording sessions, totaling **35,509 frames** with per-frame camera calibration (COLMAP), foreground masks, optional color correction LUTs, and per-frame 3D Gaussian Splatting reconstructions (see [Extract 3DGS](#extract-3dgs)).
 ## Dataset Summary
 | Session | Sequences (available / total) | Cameras | Frames (available / total) | Size |
 |---------|-------------------------------|---------|---------------------------|------|
+| s4 | 9 / 9 | 29 | 6,786 / 6,786 | ~30 GB |
+| s5 | 14 / 23 | 29 | 11,253 / 17,999 | ~49 GB |
+| s6 | 6 / 15 | 28 | 4,134 / 10,724 | ~18 GB |
+| **Total** | **29 / 47** | | **22,173 / 35,509** | **~97 GB** |
 ## Data Format
 - **Foreground masks**: `masks.tar.zst` binary person segmentation masks generated with SAM3 and with manual quality review
 - **Manifest**: `manifest.json` frame IDs, camera lists, and sequence metadata
+<details>
+<summary><b>Directory Structure</b></summary>
 ```
 DanceNet3D/
     └── ...
 ```
+</details>
+<details>
+<summary><b>Extracted Frame Structure</b></summary>
 After running `video_to_images.py`, the data is organized per-frame:
         └── ...
 ```
+</details>
 ## Quick Start
+### Extract Frames
+**Prerequisites:** Python 3.8+, FFmpeg, zstd
 ```bash
 pip install -r requirements.txt
 ```
 ```bash
 # Extract a single session
 python video_to_images.py --input s5 --output extracted/s5
 python video_to_images.py --input s5 --output extracted/s5 --apply-lut
 ```
+### Extract 3DGS
+**Prerequisites:** unzip
+Pre-computed 3D Gaussian Splatting (3DGS) reconstructions are available under the `/3dgs` directory. Each `.zip` file corresponds to one dance sequence and contains multiple `.ply` files — one per frame.
+```bash
+unzip AttitudePromenade.zip
+```
+On Mac, you can also double-click the `.zip` file in Finder. On Windows, right-click and use a tool like 7-Zip or WinRAR.
+**View a single frame:**
+1. Go to [https://superspl.at/editor](https://superspl.at/editor)
+2. Upload a single `.ply` file
+**Playback multi-frame animation:**
+1. Go to [https://superspl.at/editor](https://superspl.at/editor)
+2. Upload all `.ply` files from the sequence at once
+3. Hit the **Play** button above the timeline
 ### Using with COLMAP
 The `colmap/` directory in each sequence contains pre-computed camera intrinsics and extrinsics in COLMAP format. Camera parameters correspond to the undistorted, rotated (portrait orientation) images.
 ## Sequences
+<details>
+<summary><b>Session 4 (s4) — 9 sequences, 29 cameras</b></summary>
 | Sequence | Frames | Cameras | Status |
 |----------|--------|---------|--------|
 | ToeTaps | 572 | 29 | Available |
 | WuTang | 884 | 29 | Available |
+</details>
+<details>
+<summary><b>Session 5 (s5) — 23 sequences, 29 cameras</b></summary>
 | Sequence | Frames | Cameras | Status |
 |----------|--------|---------|--------|
 | Turns | 733 | 29 | Coming soon |
 | Twists | 930 | 29 | Coming soon |
+</details>
+<details>
+<summary><b>Session 6 (s6) — 15 sequences, 28 cameras</b></summary>
 | Sequence | Frames | Cameras | Status |
 |----------|--------|---------|--------|
 | BiancaGolden_SyncopatedGroove | 919 | 28 | Coming soon |
 | RobertRubama_RussiaRehearsal | 727 | 28 | Coming soon |
+</details>
 ## Technical Details

color_lut/0028.cube CHANGED Viewed