🎯 Quick Tips
Useful recommendations for working efficiently with the program.
🚀 For a quick start:
- Select a small folder and the pHash algorithm with 30 frames.
- Set the similarity threshold to 0.7.
- Start the scan and look at the results.
- Mark 1‑2 files and delete them (to the recycle bin) to understand the process.
📊 For accurate duplicate search:
- Use pHash with 50‑100 frames and a neighborhood of 3‑5.
- Keep the threshold at 0.7.
- Scan one large folder at a time.
- Add system folders (Windows, Program Files) to the ignore list.
🧠 For maximum accuracy (finding modified versions):
- Use CNN+Faiss with 100 frames and a neighborhood of 3.
- Set the threshold to 0.85.
- The resnet18 model – a good balance of speed and accuracy.
- Be prepared for a longer scan.
⚙️ Speeding up the scan:
- Enable pair pre‑filtering (⚙️ button) – this will discard obviously unsuitable pairs by size/duration/resolution.
- Reduce the number of frames (e.g., to 30 for pHash or 50 for CNN).
- Disable crop black borders – this will speed up frame extraction.
- Use the ignore list for folders you don’t want to scan.
- Do not add too many folders at once – it is better to scan them one by one.
🗂️ Workflow organisation:
- Use search in the results to quickly find a file by name.
- Pagination helps avoid overloading the interface when there are many pairs.
- Before bulk deletion view the pair in synchronous mode or compare in detail.
🔍 How to choose which file to keep:
- Keep the file with the larger size – this usually means higher quality.
- Compare resolution (higher is better).
- Take duration into account – a fuller version or uncropped one.
- Pay attention to FPS – for the same content, higher FPS may mean smoother playback.
- The file name can also hint at which version is original.
💾 Saving space:
Average savings after cleanup:
- Home video collection: 10‑30%.
- "Downloads" folder: 20‑50% (many repeated videos).
- Organised library: 5‑15%.
💡 Golden rule: Better to double‑check than to delete something you need. Use synchronous view for visual confirmation.