Tensorflow Object Detection Shell

Prerequisites

• python
• pip

Setup The Shell

1. Create a python virtual environment

python -m venv env

2. Activate the python virtual environment

# mac
source env/bin/activate

# windows
.\env\Scripts\activate

3. Install requirements

pip install -r requirements.txt

4. Run setup.py

python setup.py

5. Run the shell

python shell.py

Train a model

1. Collect images for training

# Using all options
>>> collect --labels <label1> <label2> --limit <limit> --folder <images_folder>

# Using minimal options (limit=5, folder=collectedimages)
>>> collect --labels <label1> <label2>

2. Label collected images

# Using all options
>>> label --folder <images_folder>

# Using minimal options (folder=collectedimages)
>>> label

3. (Optional) Load a pretrained model

>>> load --url <pretrained_model_url>

• The shell already has ssd_mobilenet_v2_fpnlite_320x320_coco17_tpu-8
• You can get more pretrained models from Tensorflow 2 Detection Model Zoo

4. Train a model

# Using all options
>>> train --model <your_model_name> --labels <label1> <label2> --sample <sample_size> --pretrained <pretrained_model_name> --steps <number_of_steps> --folder <images_folder>

# Using minimal options (pretrained=ssd_mobilenet_v2_fpnlite_320x320_coco17_tpu-8, steps=2000m, folder=collectedimages )
>>> train --model <your_model_name> --labels <label1> <label2> --train <number_of_trainees>

5. Test trained model

# Test in realtime (via webcam)
>>> test --model <your_model_name>