We are releasing a complete truly end-to-end pipeline for report generation , which you don’t have to do anything, it subscribes to the camera topics itself and does all the job
Subscribes to all the topics and runs a dinov2 model on the frames. If there are any anomalies detected, then the images get saved to the folder
A deduplication script is run, which reduces the number of duplicate images by around 100x
You can then run the Qwen model in your local machine with ollama, which would make the PDF
And voild you’ll have the PDF!
In summary, we use
- A dinov2 model for anomaly detection
- A deuplication remover that significantly reduces the images and lets you make the PDF very quickly with Qwen (which is run locally)
How does this compare with other the generators shared before us?
- We share the entire pipeline, which assumes that you just have the topics for the camera feeds, and it will work on that
- We use local models for the entire thing, from anomaly detection to even report generation
Aryan Chaudhary
CRISS Robotics