feat: crop masks with bbox and enhance postprocess efficiency

[CuriousDolphin]

Optimize Segmentation Mask Storage and Add Bounding Box Information to inference result

CONTEXT

The segmentation models were storing masks at full image resolution, leading to unnecessary memory usage and larger payload sizes in API responses or base64 encoding. Additionally, semantic and instance segmentation outputs lacked bounding box information (easily derivable from binary masks), limiting their usefulness in downstream applications.

KEY CHANGES

• Crop segmentation masks to object bounding boxes before base64 encoding
• Add bounding box coordinates to semantic and instance segmentation outputs
• Optimize mask processing with OpenCV for better performance
• Rename vision utility functions for better clarity
• Add mask_to_xyxy utility function for bbox extraction from masks
• Add Integer bbox coordinates insteaf of float

IMPACT

• Reduces API response payload size for segmentation results and latency
• Affects all segmentation model outputs (both semantic and instance)
• Improves memory efficiency in mask storage and processing
• Provides more complete object information in segmentation results

[github-actions]

Coverage Report •

File	Stmts	Miss	Cover	Missing
__init__.py	10	0	100%
local_model.py	63	0	100%
ports.py	249	8	96%	454–456, 458, 460–462, 464
remote_model.py	140	36	74%	227–229, 234, 238–239, 241–242, 246–247, 275–283, 285–288, 292–295, 299–304, 306–307, 364
runtime.py	173	97	43%	32, 40–41, 145, 149, 153, 163, 165–166, 168–170, 172, 175–177, 180, 183–185, 188, 191–192, 195–196, 199, 224–228, 230–231, 234–237, 239–240, 242, 246–249, 253–254, 256–258, 260–264, 266–267, 269–271, 275, 285–286, 296–300, 302, 304–305, 307–313, 317–320, 324–325, 327–328, 330–334, 336–337, 339–340, 342, 352–353
utils
vision.py	96	4	95%	54, 126, 164, 194
TOTAL	1069	152	85%

Tests	Skipped	Failures	Errors	Time
116	1 💤	0 ❌	0 🔥	6.644s ⏱️

[giuseppeambrosio97]

LGTM!