VAD Release Note

v1.2.0

July. 16, 2024

New Feature

• Develop percentile auto-selection function according to NG ratio of validation data
• Add overkill_intensity parameter

Improvements

• Reducing running time of inference pipeline.

Compatibility: ALO v2.4.2

v1.1.0

July. 16, 2024

New Feature

• Add image augmnetation function: brightness, contrast, saturation, hue, rotation, blur, grayscale

Compatibility: ALO v2.4.2

v1.0.0

June. 11, 2024

New Feature

• Introducing Vision Anomaly Detection(VAD) for image anomaly detection, which can select state-of-the-art PyTorch models, FastFlow and PatchCore.
• Trained exclusively on normal images, enabling effective anomaly detection without the need for abnormal image labels.
• According to infrastructure, you can run it on either CPU or GPU.

Bug Fixes

• Unify label for abnormal data as 'abnormal'

Compatibility: ALO v2.4.2