Fix process_video GPU OOM; add downsampling and dynamic GPU duration

Cut ZeroGPU quota cost of process_image via cached CPU loader

Rewrite MCP tool docstrings so schema survives Gradio's parser

Fix multi-object video inference crash by seeding all masks in one call

Accept list-form bboxes and points in image inference

Replace samv2 package with transformers SAM2

trying to fix build error

updated docs for mcp server

removed visualization for debugging

fixed bug when returning masks for multiple objects

added new variable for reference_frame_idx

increase process_video duration to 5 min

added control for dropping masks from result JSON

updated to return b64 mask decoded into string

trying async frame load

testing video inference

renamed api endpoints

fixed bug in image inferece

added interface for video

renamed image and video inference functions

added video inference imports

added input validation

trying to debug issue with F.scaled_dot_product_attention

trying to debug issue with F.scaled_dot_product_attention

testing optional input

testing optional input

added missing cv2 requirement and new function to download model weights

added in missing requirements

init comit

initial commit