SenBen: Sensitive Scene Graphs for Explainable Content Moderation
Paper • 2604.08819 • Published • 1
AI & ML interests
Ultralytics loves pushing AI forward with our community! 🚀 From YOLOv5 to YOLO26 and beyond, we make AI fast, accurate, and easy. Got ideas? Come, join us! 💡🌍
Recent Activity
View all activity
fcakyon
authored a
paper about 13 hours ago
Post
34
Let me introduce you to our CVPR 2026 paper!
Today's content moderation systems give you a label: safe or unsafe. They don't tell you what triggered the decision, who is involved, or where in the image it happens. That opacity hurts auditing, breaks adaptation across platforms, and frustrates the human review that responsible deployment demands.
We built SenBen to fix this: the first large-scale scene graph benchmark designed specifically for sensitive content moderation:
- 13,999 annotated frames from 157 movies
- Visual Genome style scene graphs with bounding boxes, attributes, and predicates
- Affective state attributes (pain, fear, aggression, distress) so the model captures not just what is in the frame, but what it means
- 16 safety tags across 5 categories, the broadest taxonomy of any dataset of this kind
A small model that beats much bigger ones:
We distilled a frontier VLM into a compact 241M parameter student built on Florence-2.
On grounded scene graph metrics, the 241M student beats every evaluated VLM except Gemini, and every commercial safety API. It also wins on object detection and captioning across the entire model zoo. It runs at 733 ms per frame on 1.2 GB VRAM, which is 7.6 times faster than the next-best local VLM at zero per-frame cost. The whole benchmark, from dataset creation through all baseline evaluations, is reproducible for under $350.
Project: https://senben.kim/
Paper: SenBen: Sensitive Scene Graphs for Explainable Content Moderation (2604.08819)
Dataset: fcakyon/senben
Code (soon): https://github.com/fcakyon/senben
Today's content moderation systems give you a label: safe or unsafe. They don't tell you what triggered the decision, who is involved, or where in the image it happens. That opacity hurts auditing, breaks adaptation across platforms, and frustrates the human review that responsible deployment demands.
We built SenBen to fix this: the first large-scale scene graph benchmark designed specifically for sensitive content moderation:
- 13,999 annotated frames from 157 movies
- Visual Genome style scene graphs with bounding boxes, attributes, and predicates
- Affective state attributes (pain, fear, aggression, distress) so the model captures not just what is in the frame, but what it means
- 16 safety tags across 5 categories, the broadest taxonomy of any dataset of this kind
A small model that beats much bigger ones:
We distilled a frontier VLM into a compact 241M parameter student built on Florence-2.
On grounded scene graph metrics, the 241M student beats every evaluated VLM except Gemini, and every commercial safety API. It also wins on object detection and captioning across the entire model zoo. It runs at 733 ms per frame on 1.2 GB VRAM, which is 7.6 times faster than the next-best local VLM at zero per-frame cost. The whole benchmark, from dataset creation through all baseline evaluations, is reproducible for under $350.
Project: https://senben.kim/
Paper: SenBen: Sensitive Scene Graphs for Explainable Content Moderation (2604.08819)
Dataset: fcakyon/senben
Code (soon): https://github.com/fcakyon/senben
onuralpszr
updated 2
Spaces 3 months ago
onuralpszr
updated a
model 3 months ago
onuralpszr
published a
Space 3 months ago
refactor: ♻️ update gradio add video and webcam and update styling of gradio
#1 opened 3 months ago
by 
onuralpszr
Update content to reflect `https://github.com/ultralytics/ultralytics/blob/main/README.md`
1
#1 opened 3 months ago
by 
lakshanthad
Update content to reflect `https://github.com/ultralytics/ultralytics/blob/main/README.md`
1
#1 opened 3 months ago
by lakshanthad
onuralpszr
published a
model 3 months ago
onuralpszr
updated a
model 3 months ago
ukltra
#4 opened 8 months ago
by 
sdsd456457541
glenn-jocher
published a
model 4 months ago
fcakyon
updated a
collection 9 months ago
onuralpszr
updated a
dataset 9 months ago
Post
3168
🎉 GitHub selected the ultralytics computer vision project, known for its YOLOv8/YOLO11 real-time SOTA computer vision models, as one of the top 5 open-source projects for first-time contributors in 2024!
Link to the project: https://github.com/ultralytics/ultralytics
Link to the full GitHub 2024 recap report: https://github.blog/news-insights/octoverse/octoverse-2024/
Link to the project: https://github.com/ultralytics/ultralytics
Link to the full GitHub 2024 recap report: https://github.blog/news-insights/octoverse/octoverse-2024/
- 2 replies