Table 1 Augmentation operations.

From: TEMImageNet training library and AtomSegNet deep-learning models for high-precision atom segmentation, localization, denoising, and deblurring of atomic-resolution images

Operations

Randomization

Field of view

8, 10, 20, 30, 40 angstroms

Rotation

0°–90° with 15° intervals

Background

Constant, linear ramp, non-linear patterns

Shot and scan noise

The shot and scan noise level is randomized

Nanoparticle shape

The distance of the 111 facets have been randomized in the dataset

Position offset

Randomized

Imaging conditions

1. 200 keV, 24 mrad, source size = 0.9 Å, C3/5 = 0, df = 0, \({\sigma }_{jitter}=0.2\) Å

2. 100 keV, 30 mrad, source size = 0.8 Å, C3/5 = 0, df = 0, \({\sigma }_{jitter}=0.2\) Å

3. 200 keV, 10.5 mrad, source size = 0.9 Å, C3/5 = 0, df = 0, \({\sigma }_{jitter}=0.2\mathrm{ \AA }\)

4. 200 keV, 10.5 mrad, source size = 0.9 Å, C3/5 = 0, df = 0, \({\sigma }_{jitter}=0.2\mathrm{ \AA }\)

5. 200 keV, 10 mrad, source size = 1.6 Å, C3/5 = 0, df = 0, \({\sigma }_{jitter}=0.2\mathrm{ \AA }\)