Savant
Savant copied to clipboard
Python Computer Vision & Video Analytics Framework With Batteries Included
When you experiment with different input sizes you want keep engines for them instead of being replaced with the engines generated when the model is built with a new input...
Explore the feasibility of implementing a universal converter for standard YOLO-based KP models.
The docs: https://docs.nvidia.com/metropolis/deepstream/dev-guide/text/DS_plugin_gst-nvtracker.html#shadow-tracking-target-data
We meet situations when we want to not scale a frame to the model input but use tiling, when we cover the frame surface with ROIs, matching the model input...
https://github.com/insight-platform/Savant/blob/21bc88dbfa5643e99dcf6618c4dd8f7495929337/savant/parameter_storage/__init__.py#L8