Tensorflow add-on summary box

Introduction

Add-on implementation for Tensorflow object detection. Summary table will be displayed at the right top corner of the inference image. Also, this add-on enable to save details of detection data (class, score(%), xmin, ymin, xmax, ymax) in csv format on the inference run. The fontsize can either be assigned manually in the code (visualization_utils.py) or adjusted automatically based on the screen width (see line :1128). This add-on will be useful to capture the summarized overview of the object detection at a glance. 😼

How to Apply

1. After cloning this repo, replace the following files.

• models/research/object_detection/utils/visualization_utils.py
• models/research/object_detection/Object_detection_image_tf2.py

2. install & reload dependencies

pip install tabulate=0.9.0

cd models/research/

python -m pip install .

How to run

Configure Object_detection_image_tf2.py file before run.

# Number of classes the object detector can identify
NUM_CLASSES = 20

# IMAGE_NAME
IMAGE_NAME = 'targe_image1.jpg'
# IMAGE_SAVE_NAME
IMAGE_SAVE_NAME = 'targe_image1_result'

# Path to frozen detection graph .pb file, which contains the model that is used
# for object detection.
PATH_TO_PB = '/PATH/TO/BE/CONFIGURED/PB_folder/saved_model'
# Path to label map file
PATH_TO_LABELS = '/PATH/TO/BE/CONFIGURED/labelmap.pbtxt'
# Path to image
PATH_TO_IMAGE_DIR = '/PATH/TO/test_dir'
PATH_TO_IMAGE = os.path.join(PATH_TO_IMAGE_DIR,IMAGE_NAME)

Run Object_detection_image_tf2.py under models/research/object_detection/

python Object_detection_image_tf2.py

If you want to specify GPU

export CUDA_VISIBLE_DEVICES=0 && python Object_detection_image_tf2.py

If you want to adjust the table format

Edit header at Line:1140

  headers=["Class              ", "total%", "  Qty  "]

Or edit multiplier of auto fontsize

    Summary_FontSIZE = int(im_width * 0.05)

if you want to change font color and size

Edit values at Line:47~50 of visualization_utils.py

# Set config for summary box (HM)
FontSIZE = 10
Auto_Fontsize=True
Summary_FontSIZE = 20
FontColor = 'white'

Debug

• If font size is not working, that might be case matter.. "a"rial

# use whatever works
font = ImageFont.truetype('Arial.ttf', Summary_FontSIZE)
font = ImageFont.truetype(‘arial.ttf', Summary_FontSIZE)

Reference

• https://pypi.org/project/tabulate/
• Tensorflow base: https://github.com/tensorflow/models/tree/master/research/object_detection