Detection Visualization

This projects uses the tensorflow object detection utilities to draw bounding boxes, detection scores and object labels on a frame.

The result of each Visualization are then saved as a part of a sequence of images in an output folder.

It implements a service that consumes detection output from https://github.com/kunadawa/video-object-detection via grpc.

Setup

• download or clone this repo
• Download or clone the tensorflow models repo
• Download or clone the video object detection repo
• In this repo's root, make the following soft links so that the visualization utils imports can work
• [tensorflow-models-root]/research/object_detection
• [video object detection repo]/proto
• add [video object detection repo]/proto/generated to PYTHONPATH

Running

• python visualize_image.py object_detection/data/mscoco_complete_label_map.pbtxt 50001

Replace the label map path as appropriate

Tests

visualize_image_tests.py uses a shared fixture defined in [video object detection repo]/samples/conftest.py. Create a symlink to that file to the root directory then run pytest visualize_image_tests.py