dataku / google_colabs /README.md
ahmad walidurosyad
add
9fd445b

Google Colab Gradio Interface

This folder contains two versions of the Google Colab notebook:

Files

1. Video_Subtitle_Remover_Gradio.ipynbNEW - Recommended

Gradio Web Interface - Easy-to-use browser-based UI

Features:

  • 🖱️ Click-and-upload interface (no coding required)
  • 🎨 Visual algorithm selection
  • ⚙️ Adjustable parameters with sliders
  • 📊 Real-time progress tracking
  • 📥 One-click download

Best for:

  • Users who prefer GUI
  • Quick testing
  • Non-technical users
  • Multiple video processing

Usage:

  1. Open in Colab
  2. Run all cells
  3. Click the generated link
  4. Use web interface in browser

2. Video_Subtitle_Remover.ipynb

Traditional Notebook - Code-based approach

Features:

  • Step-by-step execution
  • Full control over parameters
  • Good for understanding the process
  • Batch processing scripts

Best for:

  • Users comfortable with code
  • Custom workflows
  • Debugging
  • Learning the internals

Quick Start

For Gradio Interface (Recommended):

1. Open Video_Subtitle_Remover_Gradio.ipynb in Colab
2. Runtime → Change runtime type → GPU
3. Run all cells (Ctrl+F9)
4. Click the gradio.live URL
5. Upload video and click "Remove Subtitles"

For Traditional Notebook:

1. Open Video_Subtitle_Remover.ipynb in Colab
2. Runtime → Change runtime type → GPU
3. Run cells step by step
4. Configure settings in Step 5
5. Run processing in Step 7

Algorithm Recommendations

Use Case Algorithm Quality Speed
Best Quality DiffuEraser ⭐⭐⭐⭐⭐ ⭐⭐
Fastest STTN ⭐⭐⭐ ⭐⭐⭐⭐⭐
Balanced Stable Diffusion ⭐⭐⭐⭐ ⭐⭐⭐
High Motion ProPainter ⭐⭐⭐⭐⭐

System Requirements

  • GPU: Required (T4/P100/V100)
  • Storage: 10-20GB for models
  • VRAM:
    • STTN: 4GB
    • DiffuEraser: 12GB
    • Stable Diffusion: 8GB

Performance (Colab T4 GPU)

Video Algorithm Time
1 min 720p STTN ~30s
1 min 720p DiffuEraser ~3-5min
5 min 720p STTN ~2min
5 min 720p DiffuEraser ~15-20min

Troubleshooting

Gradio not loading

  • Wait 30-60 seconds for models to load
  • Check if all cells ran successfully
  • Restart runtime and try again

Out of Memory

  • Reduce batch size in settings
  • Use STTN instead of DiffuEraser
  • Process shorter videos

Slow processing

  • Use STTN for preview
  • Enable GPU in Colab settings
  • Consider Colab Pro for unlimited runtime

Links

Tips

  1. Start with STTN to test quickly
  2. Use DiffuEraser for final high-quality output
  3. Keep videos under 10 minutes on free tier
  4. Save to Google Drive to avoid data loss
  5. Monitor GPU usage with !nvidia-smi