NASA MicroNet

← Back to Image Segmentation

NASA’s pretrained microscopy models (MicroNet) provide convolutional neural networks trained on a diverse collection of materials microscopy images. The pretrained features transfer well to new microscopy domains, often outperforming ImageNet-pretrained initializations when fine-tuned on domain-specific data.

Reference: https://github.com/nasa/pretrained-microscopy-models

See also

• U-Net — the encoder-decoder architecture MicroNet backbones plug into

• Segment Anything Model (SAM v1) — a general-purpose (non-microscopy-specific) foundation model alternative

• Image Segmentation — every other segmentation and tracking approach