Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,54 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: cc-by-nc-sa-4.0
|
| 3 |
+
task_categories:
|
| 4 |
+
- video-inpainting
|
| 5 |
+
- watermark-removal
|
| 6 |
+
- object-removal
|
| 7 |
+
language:
|
| 8 |
+
- en
|
| 9 |
+
tags:
|
| 10 |
+
- capcut
|
| 11 |
+
- ugc-video
|
| 12 |
+
- benchmark
|
| 13 |
+
- deep-learning
|
| 14 |
+
pipeline_tag: video-to-video
|
| 15 |
+
---
|
| 16 |
+
|
| 17 |
+
# CapCut Video Watermark Benchmark Dataset
|
| 18 |
+
|
| 19 |
+
## Dataset Description
|
| 20 |
+
|
| 21 |
+
This repository contains a specialized **benchmark dataset** for evaluating the performance of video inpainting and watermark removal models, specifically focusing on content generated by **CapCut (剪映)**, a popular mobile video editing platform.
|
| 22 |
+
|
| 23 |
+
CapCut videos often feature distinct watermark patterns, including:
|
| 24 |
+
|
| 25 |
+
* **Static Corner Logos:** Opaque or semi-transparent logos (e.g., "CapCut") in a fixed corner.
|
| 26 |
+
* **Dynamic End Screens:** A full-screen animation with the CapCut logo at the very end of the video.
|
| 27 |
+
* **Overlay Watermarks:** Sometimes a username or ID is overlaid on the video content.
|
| 28 |
+
|
| 29 |
+
### Dataset Purpose
|
| 30 |
+
* **Benchmarking:** Provide a standardized evaluation set for comparing different video inpainting algorithms on CapCut-specific artifacts.
|
| 31 |
+
* **Research:** Analyze the unique challenges posed by CapCut's watermarking techniques.
|
| 32 |
+
* **Training Data:** Serve as a fine-tuning dataset for models aiming to handle UGC video content.
|
| 33 |
+
|
| 34 |
+
### Data Structure & Access
|
| 35 |
+
|
| 36 |
+
The dataset includes:
|
| 37 |
+
* **Original (Watermarked):** Short video clips (5-15 seconds) with CapCut watermarks.
|
| 38 |
+
* **Ground Truth (Clean):** Corresponding clean, watermark-free versions of the videos.
|
| 39 |
+
* **Masks:** Binary segmentation masks indicating the location of the watermark in each frame.
|
| 40 |
+
|
| 41 |
+
**To access the dataset for research purposes, please request access through the Hugging Face platform.**
|
| 42 |
+
|
| 43 |
+
## 🏆 Benchmark Results & Optimized Tool
|
| 44 |
+
|
| 45 |
+
We have extensively tested our own **advanced video inpainting models** on this dataset. The model achieving the highest performance in terms of both PSNR/SSIM metrics and perceptual quality for CapCut content is our specialized, online tool.
|
| 46 |
+
|
| 47 |
+
For users seeking the **most effective, one-click solution for removing CapCut watermarks** based on these benchmark results:
|
| 48 |
+
|
| 49 |
+
👉 **[Try the CapCut Watermark Remover (Optimized)](https://www.videowatermarkremove.com/remove-capcut-watermark)**
|
| 50 |
+
|
| 51 |
+
(Our online platform integrates specific loss functions for CapCut logos and end screens, running on high-performance GPUs.)
|
| 52 |
+
|
| 53 |
+
---
|
| 54 |
+
*Disclaimer: This dataset is provided for academic research and benchmarking purposes only. Please respect copyright laws when using the data.*
|