Update README.md
Browse files
README.md
CHANGED
|
@@ -101,3 +101,40 @@ print("Predictions:")
|
|
| 101 |
for i, prob in enumerate(verdict.predictions):
|
| 102 |
print(f" Label {i}: {prob * 100:.2f}%")
|
| 103 |
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 101 |
for i, prob in enumerate(verdict.predictions):
|
| 102 |
print(f" Label {i}: {prob * 100:.2f}%")
|
| 103 |
```
|
| 104 |
+
|
| 105 |
+
## Labels
|
| 106 |
+
|
| 107 |
+
The model can detect the following labels:
|
| 108 |
+
|
| 109 |
+
- **AI_GEN**: Is the video AI-generated or not?
|
| 110 |
+
- **ANIME_1D**: Is the video in 2D anime style?
|
| 111 |
+
- **ANIME_2D**: Is the video in 3D anime style?
|
| 112 |
+
- **VIDEO_GAME**: Does the video look like a video game?
|
| 113 |
+
- **KLING**: Is the video generated by Kling?
|
| 114 |
+
- **HIGGSFIELD**: Is the video generated by Higgsfield?
|
| 115 |
+
- **WAN**: Is the video generated by Wan?
|
| 116 |
+
- **MIDJOURNEY**: Is the video generated using images from Midjourney?
|
| 117 |
+
- **HAILUO**: Is the video generated by Hailuo?
|
| 118 |
+
- **RAY**: Is the video generated by Ray?
|
| 119 |
+
- **VEO**: Is the video generated by Veo?
|
| 120 |
+
- **RUNWAY**: Is the video generated by Runway?
|
| 121 |
+
- **SORA**: Is the video generated by Sora?
|
| 122 |
+
- **CHATGPT**: Is the video generated using images from ChatGPT?
|
| 123 |
+
- **PIKA**: Is the video generated by Pika?
|
| 124 |
+
- **HUNYUAN**: Is the video generated by Hunyuan?
|
| 125 |
+
- **VIDU**: Is the video generated by Vidu?
|
| 126 |
+
|
| 127 |
+
> **Note**: The **AI_GEN** label is the most accurate as it has the most training data. Other labels have limited training data and may be less accurate.
|
| 128 |
+
|
| 129 |
+
## Accuracy
|
| 130 |
+
|
| 131 |
+
The PR curve of the model is shown below:
|
| 132 |
+
|
| 133 |
+
<p align="center">
|
| 134 |
+
<img src="https://github.com/LaunchPlatform/cakelens-v5/raw/master/assets/pr-curve.png?raw=true" alt="PR Curve" />
|
| 135 |
+
</p>
|
| 136 |
+
|
| 137 |
+
At threshold 0.5, the model has an precision of 0.77 and a recall of 0.74.
|
| 138 |
+
The dataset size is 5,093 videos for training and 498 videos for validation.
|
| 139 |
+
Please note that the model is not perfect and may make mistakes.
|
| 140 |
+
|