Xuweiyi/pre-release-dre10k

Dynamic RE10K Dataset - Split Archives

This directory contains the Dynamic RE10K dataset split into multiple archives for easier download and upload.

📊 Dataset Overview

• Dataset: Dynamic RE10K (Real Estate 10K with dynamic scenes)
• Total Scenes: ~8,000 clips
• Train/Test Split: Included
• Format: PNG images + JSON metadata + MP4 videos
• Archive Format: tar.gz (gzip compressed)
• Chunk Size: ~2 GB per archive

📦 Files in This Directory

dynamic_re10k_part001.tar.gz    # Archive chunk 1
dynamic_re10k_part002.tar.gz    # Archive chunk 2
...
dynamic_re10k_partXXX.tar.gz    # Archive chunk N
dynamic_re10k_manifest.json     # Manifest with checksums (IMPORTANT!)

⚠️ IMPORTANT: Download ALL files including dynamic_re10k_manifest.json

---

🔓 Extraction Methods

Method 1: Automated Extraction with Verification (Recommended)

Prerequisites: Python 3.7+

Download the extraction script:

wget https://raw.githubusercontent.com/YOUR_REPO/scripts/extract_split_archives.py

Extract and verify:

python extract_split_archives.py \
  --archives-dir . \
  --output-dir ./dynamic_re10k \
  --manifest ./dynamic_re10k_manifest.json

What it does:

• ✅ Verifies MD5 checksums for all archives
• ✅ Detects missing or corrupted files
• ✅ Extracts all archives to output directory
• ✅ Validates extraction

---

Method 2: Manual Extraction

Extract all archives (Linux/Mac):

for f in dynamic_re10k_part*.tar.gz; do
    echo "Extracting $f..."
    tar -xzf "$f"
done

Extract all archives (Windows PowerShell):

Get-ChildItem dynamic_re10k_part*.tar.gz | ForEach-Object {
    Write-Host "Extracting $_..."
    tar -xzf $_.Name
}

Extract single archive:

tar -xzf dynamic_re10k_part001.tar.gz

---

✅ Verify Integrity (After Extraction)

Check Manifest

# View manifest
cat dynamic_re10k_manifest.json

# The manifest contains:
# - Number of chunks
# - MD5 checksums for each archive
# - File counts and sizes

Verify Checksums (Linux/Mac)

# Create checksum verification
for chunk in $(jq -r '.chunks[].filename' dynamic_re10k_manifest.json); do
    expected_md5=$(jq -r ".chunks[] | select(.filename==\"$chunk\") | .md5" dynamic_re10k_manifest.json)
    actual_md5=$(md5sum "$chunk" | cut -d' ' -f1)
    
    if [ "$expected_md5" = "$actual_md5" ]; then
        echo "✓ $chunk"
    else
        echo "✗ $chunk (checksum mismatch!)"
    fi
done

Verify Checksums (Python)

import json
import hashlib
from pathlib import Path

def verify_md5(file_path):
    hash_md5 = hashlib.md5()
    with open(file_path, "rb") as f:
        for chunk in iter(lambda: f.read(4096), b""):
            hash_md5.update(chunk)
    return hash_md5.hexdigest()

# Load manifest
with open('dynamic_re10k_manifest.json', 'r') as f:
    manifest = json.load(f)

# Verify each chunk
for chunk in manifest['chunks']:
    filename = chunk['filename']
    expected = chunk['md5']
    actual = verify_md5(filename)
    
    if expected == actual:
        print(f"✓ {filename}")
    else:
        print(f"✗ {filename} - Checksum mismatch!")

---

📁 Expected Directory Structure (After Extraction)

dynamic_re10k/
├── train/
│   ├── images/
│   │   ├── scene1_clip0000/
│   │   │   ├── 00000.png
│   │   │   ├── 00001.png
│   │   │   └── ...
│   │   ├── scene1_clip0001/
│   │   └── ...
│   ├── metadata/
│   │   ├── scene1_clip0000.json
│   │   ├── scene1_clip0001.json
│   │   └── ...
│   ├── videos/
│   │   ├── scene1_clip0000.mp4
│   │   ├── scene1_clip0001.mp4
│   │   └── ...
│   └── full_list.txt
├── test/
│   ├── images/
│   ├── metadata/
│   ├── videos/
│   └── full_list.txt
├── test_scenes_info.json
└── test_scenes.json

---

📋 Dataset Information

Train Split

• Scenes: ~3,900+ clips
• Frames per clip: 20-40 frames
• Total frames: ~160,000+
• Format: 1080x1080 PNG (center-cropped)

Test Split

• Scenes: ~77 clips
• Frames per clip: 20-40 frames
• Total frames: ~3,000+
• Format: 1080x1080 PNG (center-cropped)

Metadata Format (JSON)

Each scene has a metadata JSON file containing:

{
  "scene_name": "scene_clip0000",
  "source_video": "scene_name",
  "original_frame_range": [0, 39],
  "num_original_frames": 40,
  "num_kept_frames": 40,
  "frames": [
    {
      "frame_idx": 0,
      "original_frame_idx": 0,
      "image_path": "/absolute/path/to/00000.png",
      "fxfycxcy": [fx, fy, cx, cy],
      "w2c": [[...], [...], [...], [...]]
    },
    ...
  ]
}

---

🚀 Quick Start

# 1. Download all archives and manifest

# 2. Extract all
for f in dynamic_re10k_part*.tar.gz; do tar -xzf "$f"; done

# 3. Verify structure
ls -la dynamic_re10k/train/images/ | head -20
ls -la dynamic_re10k/test/images/ | head -20

# 4. Check train/test lists
wc -l dynamic_re10k/train/full_list.txt
wc -l dynamic_re10k/test/full_list.txt

---

💾 Storage Requirements

• Compressed (archives): ~20-30 GB
• Extracted: ~45-50 GB
• Total required (for extraction): ~70-80 GB

Make sure you have enough disk space before extraction!

---

🐛 Troubleshooting

"Archive is corrupted"

• Re-download the specific archive
• Verify MD5 checksum matches manifest
• Ensure download completed fully

"Missing files after extraction"

• Check if all archives were extracted
• Verify all archives were downloaded
• Check disk space

"tar: command not found" (Windows)

• Install Git for Windows (includes tar)
• Or use 7-Zip to extract .tar.gz files
• Or use WSL (Windows Subsystem for Linux)

"Disk space full"

• Archives + extracted need ~70-80 GB
• Delete archives after successful extraction to save space

---

📝 Citation

If you use this dataset, please cite:

---

📧 Support

For issues or questions:

• Open an issue on GitHub
• Check the Hugging Face dataset page
• Contact: xuweic@email.virginia.edu

---

📜 License

[Specify your license here]

---

Last Updated: 2025-10 Version: 1.0