In this project, we show how to detect and also segment images, based on pre-trained Tensorflow based object detection models.
Detection is identifying a bounding box, instance segmentation requires identifying the pixels that can belong to the object.
We use R-Mask CNN model with inception 2. The R-CNN is does detection in multiple stage with multiple CNN passes. At first it generated proposal regions by studying pixel gradients and edges. Classification is run on each proposal. The final results and bounding boxes are then obtained post-processing of the individual results.
Loading comments...