phronetic-ai/owlet-safety-3b-1

VISIONPricing:Input $0.32 / Cached $0.016 / Output $1.6Concurrent Unit Cost:1Model Size:3BQuant:BF16Context Size:32kPublished:Jun 17, 2025License:cc-by-4.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

phronetic-ai/owlet-safety-3b-1 is a 3 billion parameter multi-modal vision-language model, fine-tuned from Qwen2.5-VL-3B-Instruct, specifically designed for multi-label safety event detection in video clips. It identifies events like fire, smoke, fall, assault, sos, and theft, making it highly suitable for video surveillance, incident detection, and safety monitoring applications where simultaneous events need to be recognized. The model processes chat-style prompts and video inputs, outputting a comma-separated list of detected safety labels.

Loading preview...

Owlet Safety 1: Multi-Label Video Event Detection

phronetic-ai/owlet-safety-3b-1 is a specialized 3 billion parameter multi-modal vision-language model, fine-tuned from the Qwen2.5-VL-3B-Instruct base. Its primary function is to perform multi-label safety event detection within video clips, distinguishing it from general-purpose LLMs by its focused application in video analysis.

Key Capabilities

  • Multi-Label Detection: Identifies multiple safety events concurrently within a single video, such as fire, smoke, fall, assault, sos, and theft.
  • Video Input Processing: Designed to take video clips as input, alongside chat-style prompts, to analyze visual information for safety concerns.
  • Domain-Specific Fine-tuning: Optimized on domain-specific video datasets with multi-label safety annotations, ensuring high relevance and accuracy for surveillance tasks.
  • Efficient Deployment: The adapter weights have been merged into the base model, simplifying deployment and usage.

Good For

  • Video Surveillance: Automating the detection of critical safety incidents in real-time or recorded video feeds.
  • Incident Detection: Rapidly identifying specific events like fire, fall, or theft in various environments.
  • Safety Monitoring: Enhancing security systems by providing automated alerts for potential dangers or emergencies.
  • Resource-Constrained Environments: With options for int8 and int4 quantization, the model can run efficiently on GPUs with as little as 2.5 GB VRAM, making it accessible for diverse hardware setups.

Finetunes
1 models