Spaces:
Running on Zero
Running on Zero
| title: Peak-End-Net Video Aesthetic Assessment | |
| emoji: 🎬 | |
| colorFrom: green | |
| colorTo: gray | |
| sdk: gradio | |
| sdk_version: 6.20.0 | |
| app_file: app.py | |
| short_description: Video aesthetic scoring via peak-end rule (11 scores) | |
| python_version: "3.12" | |
| startup_duration_timeout: 30m | |
| # Peak-End-Net: Video Aesthetic Assessment | |
| Upload a video to get an **overall aesthetic score** and **ten fine-grained attribute scores** | |
| based on the *peak-end rule* from cognitive psychology. | |
| ## How it works | |
| Peak-End-Net evaluates video aesthetics by focusing on the most salient ("peak") moments | |
| and the ending of a video, rather than averaging all frames equally. It uses a frozen CLIP | |
| ViT-L/14 backbone with an AVA-pretrained aesthetic head, a key-moment discovery module, | |
| peak-end aggregation, a rhythm encoder, and a gated fusion module. | |
| ## Scores | |
| | Score | Description | | |
| |---|---| | |
| | **overall** | Overall aesthetic quality (gated fusion) | | |
| | **composition** | Visual composition and framing | | |
| | **shotsize** | Shot size / framing distance | | |
| | **lighting** | Lighting quality and mood | | |
| | **visualtone** | Visual tone and atmosphere | | |
| | **color** | Color palette and grading | | |
| | **depthoffield** | Depth of field and focus | | |
| | **expression** | Emotional expression (human subjects) | | |
| | **movement** | Camera and subject movement | | |
| | **costume** | Costume and styling | | |
| | **makeup** | Makeup quality | | |
| ## Model | |
| - **Model:** [GD-ML/Peak-End-Net](https://huggingface.co/GD-ML/Peak-End-Net) | |
| - **Paper:** [arXiv:2607.13941](https://arxiv.org/abs/2607.13941) | |
| - **Code:** [GitHub](https://github.com/AMAP-ML/Peak-End-Net) |