Reproduce “DCM-SAM: Defect-Conditioned Mixture of LoRA Experts for NPU-Deployed AM Defect 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
Md Mushfiqur Rahaman, Md Mahedi Hasan, Imtiaz Ahmed, Srinjoy Das
DCM-SAM adapts SAM for metal additive-manufacturing inspection by training separate, prompt-free segmentation paths for pores and inclusions while retaining a frozen ViT-B backbone and full image resolution. The complete model runs in FP16 on a Qualcomm Hexagon NPU at 1024 × 1024. It reports 64.2% pore IoU on real NIST scans despite synthetic-only training, but NIST post-processing was calibrated per split, one specimen had special processing, and the baseline comparison was not protocol-matched. The attention rewrite enables deployment, not a demonstrated speed advantage, and latency at that resolution was not measured.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.