Instance Segmentation Pipeline - Mask R-CNN, PyTorch
Easily create custom computer vision models to detect and mask objects in your images and video for counting, movement, shape, and proportions.
- Frame Reduction and Proposal - From a video, retain only the frames consisting of the visual content you're interested in, using the Cognitive Services, Scene and Activity Recognition Service.
- Frame Annotation - Annotate frames using VGG Image Annotator, COCO Annotation UI, Labelbox or another image annotation tool.
- Visualise Class Distribution - View the accumulation of samples per class. Extend this notebook to describe frames using custom metadata for subsetting.
- Prepare Training Dataset - Create Training and Validation sets using metadata to subset.
- View Training Dataset - View the Training and Validation sets - and right any wrongs from annotating frames - prior to model training.
- Model Training - Let the magic begin. Be sure to specify the Training and Validation set and initial model weights, and then experiment with different training regimes.
- Model Scoring