TensorFlOW

A collection of trained machine learning models to detect HP bars in the game Overwatch. Images were all collected from the game myself.

Current trained models:

• faster_rcnn_inception_v2
• ssd_mobilenet_v1

Getting Started

Dependencies

• tensorflow-gpu
• cv2
• mss
• numpy

Folder Structure

+ detection examples: contains example images of detected health bars using the given models
+ images: contains train/test images used to train the models
+ models: contains files of the different models (also has the trained inference graphs)
+ utils: scripts used to create the tfrecords and labels (Credit: https://github.com/datitran/raccoon_dataset)
- file_detection.py: Shows and saves detections from a folder of images
- live_detection.py: Shows detections from screenshots of the screen

Models

All models were trained using the TensorFlow object detection API. FPS was measured on a system with an i7-7700k and a GTX 1080.

faster_rcnn_inception_v2

Avg. detection FPS: 12

MIN_SCORE_THRESH: 0.97

Example 1:

Example 2:

Example 3:

ssd_mobilenet_v1

Avg. detection FPS: 52

MIN_SCORE_THRESH: 0.6

Example 1:

Example 2:

Example 3: