Fig. 2: Performance evaluation of LiteLoc on SMLM data with different PSFs. | Nature Communications

Fig. 2: Performance evaluation of LiteLoc on SMLM data with different PSFs.

From: Scalable and lightweight deep learning for efficient high accuracy single-molecule localization microscopy

Fig. 2

a Analysis speed of LiteLoc, DECODE, and DeepSTORM3D running on a single GPU. Insets are model size (number of parameters) and computational workload. b LiteLoc demonstrates near-linear scalability of analysis speed on multiple GPUs. All tests were conducted on NVIDIA GeForce RTX 4090 GPUs. c, d Localization accuracy for astigmatic PSF and 6 µm DMO-tetrapod PSF, respectively. The blue line for \({{{CRLB}}_{x}}^{1/2}\) is underneath the green line for \({{CRLB}}_{y}^{1/2}\). 5000 photons and 50 background photons were used for each single molecule. 2000 single-emitter images with random x, y positions were generated for each axial position. e, f Performance evaluation on simulated datasets with different densities and SNRs using 3D Efficiency and RMSE based on astigmatic PSF and 6 µm DMO-tetrapod PSF, respectively. Scale bar, 1 µm (c, d).

Back to article page