Reproduce “PicoSAM3: Real-Time In-Sensor Region-of-Interest Segmentation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Pietro Bonazzi, Nicola Farronato, Stefan Zihlmann, Haotong Qin, Michele Magno
PicoSAM3 enables object-mask prediction on the IMX500 sensor by using a bounding box to crop the image, then feeding the cropped RGB pixels to a compact convolutional network without explicit prompt inputs. The INT8-quantized Q-PicoSAM3 reports 65.34% COCO mIoU and 63.98% LVIS mIoU, with 1.3 MB size and 11.82 ms inference latency; interactive ROI updates remain rate-limited to 150 ms.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.