image_instance_segmentation

README

Image Instance Segmentation

The example uses mask_rcnn_inception_v2_coco trained on the COCO dataset.

Refer to Tensorflow detection model zoo for more details.

The input and output nodes of the model

Node Name	Input/Output	Shape	Data Description
image_tensor	Input	[batch, height, width, 3]	RGB pixel values as uint8 in a square format (Width, Height).
detection_boxes	Output	[batch, num_detections, 4]	Array of boxes for each detected object in the format [yMin, xMin, yMax, xMax]
detection_scores	Output	[batch, num_detections]	Array of probability scores for each detected object between 0 and 1
detection_classes	Output	[batch, num_detections]	Array of object class indices for each object detected based on COCO objects
detection_masks	Output	[batch, mask_height, mask_height]	Array of instance masks

Usage

go run main.go -dir=<model folder> -jpg=<input.jpg> [-out=<output.jpg>] [-labels=<labels.txt>]

References

• Run an Instance Segmentation Model
• TensorFlow visualization_utils