PicoSAM3: Real-Time In-Sensor Region-of-Interest Segmentation

Dimensions overall score 8.0

GitHub Code Pulse

No public code linked for this paper yet.

• PicoSAM3 achieves 65.45% mIoU on the COCO dataset.result 95%
• PicoSAM3 achieves 64.01% mIoU on the LVIS dataset.result 95%
• PicoSAM3 has 1.3 million parameters.technical 95%
• The INT8 quantized PicoSAM3 model enables real-time in-sensor inference at 11.82 ms latency on the Sony IMX500 vision sensor.technical 90%
• Knowledge distillation from large SAM models yields up to +14.5% mIoU improvement over supervised training for PicoSAM3.result 90%
• PicoSAM3 outperforms existing SAM-based and edge-oriented baselines at similar or lower complexity.result 85%
• PicoSAM3 is optimized for deployment on the Sony IMX500 vision sensor.technical 90%
• The INT8 quantized model preserves accuracy with negligible degradation.technical 85%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card