Olive-Separation-With-TensorFlow-Object-Detection-API

The task is to identify and to count black and green olives in the video.

The steps:

1. Make 75 photos.

(25 black olives, 25 green olives, 25 both)

Examples of images:

2. Transform images into a size of 300x300.

Using transform_image_resolution.py

3. Create annotations in LableImg.

Creating CSV format from XML using xml_to_csv.py

4. Generate TFRecords.

Using generate_tfrecord.py

5. Create a label map file and a configuration file.

The used configuration file is https://github.com/tensorflow/models/blob/master/research/object_detection/configs/tf2/ssd_efficientdet_d0_512x512_coco17_tpu-8.config

6. Train model.

The used model is http://download.tensorflow.org/models/object_detection/tf2/20200711/efficientdet_d0_coco17_tpu-32.tar.gz

7. Export the inference graph.

8. Test the model.

Input:

Output:

8 black olives.

1 green olives.