Real Time Object Detection for Libras

To activate the virtual environment:

source tf2_api_env/bin/activate
cd ~/Downloads/Tensorflow

To run the jupyter notebok use:

python3 -m notebook

or

jupyter notebook

To train, use:

python ~/Downloads/Tensorflow/models/research/object_detection/model_main_tf2.py --model_dir=Tensorflow/workspace/models/my_ssd_mobnet --pipeline_config_path=Tensorflow/workspace/models/my_ssd_mobnet/pipeline.config --num_train_steps=5000

Tutorial

Environment instalation tutorial: https://neptune.ai/blog/how-to-train-your-own-object-detector-using-tensorflow-object-detection-api

Implementation tutorial: https://www.youtube.com/watch?v=pDXdlXlaCco

Passo a passo:

1. Tirar as fotos que servirão para treinar a IA.

• Foram 426 fotos (16 por letra do alfabeto).

2. Rotular as fotos usando labelImg.
3. Criar um label map.
4. Treinar.