Capsule inspection

Visual inspection for the pharmaceutical industry using convolutional neural networks.

Proof of concept

Classification models

Softgel classification model:

• Model: ResNet18
• Val. accuracy: 88.23%
• Val. AUC: 0.9122

Capsule classification model:

• Model: ResNet18
• Val. accuracy: 95.43%
• Val. AUC: 0.8858

Segmentation models

Trained UNet models with a ResNet18 backbone which achieved very low loss values on both datasets.

Links

Data:

• SensumSODF Dataset

Tools:

• Weights & Biases Dashboard