Machine Washer Part Density Predictions
- Benchmarked various image thresholding methods to count highly-occluded machine parts like Otsu Thresholding, Gaussian Filtering, Sobel Filtering and Hough Circle Transforms.
- Extended density-map estimation based FamNet to count the machine parts with a novel mismatch loss component, achieving a performance of 1.96 MAE (90% decrease from image processing baseline) on the challenge dataset.