I created a project with the purpose of using Grounded-DINO, SAM (Segment Anything Model), and tracking algorithms to achieve text-prompt-based object recognition and continuous tracking in videos. This combination allows for precise and efficient identification and tracking of objects within video content based on textual descriptions.
Objectives:
1. Text-Prompt Based Object Recognition:
• Utilize Grounded-DINO to interpret and understand textual prompts for object identification within video frames.
2. Segmentation and Analysis:
• Implement SAM (Segment Anything Model) to accurately segment and analyze objects in video frames based on the prompts provided.
3. Continuous Object Tracking:
• Apply sam2 tracking algorithms to maintain and follow the identified objects throughout the video, ensuring consistent and reliable tracking over time.
Benefits:
• Efficiency: Streamline the process of object recognition and tracking by leveraging state-of-the-art models.
• Accuracy: Enhance the precision of object identification and tracking through advanced segmentation techniques.
• Automation: Enable automated monitoring and analysis of video content based on textual descriptions, reducing the need for manual intervention.
This project aims to integrate cutting-edge technologies in computer vision and natural language processing to create a robust system for video content analysis and tracking.
prepare the images data as the follow rules: /your_own_path/raw_data
where raw_data is fixed
You could use the dockerfile, and create environment as Grounded-SAM
cd /home/appuser/Grounded-Segment-Anything
Then command example:
python grounded_sam_with_sam_tracking.py -i /your_path_to_data/raw_data -o /your_path_to_data/ --box_threshold 0.23
If you wan to see the result of pretraining, please run:
python draw_raw_image_and_box.py
#Result
