- Here we have used the ssd_mobilenet_v2_fpnlite 320x320 for object detection
- The pretrained models are cloned and imported from tensorflow/models
- We use the tensorflow_gpu - 2.9.1 along with CUDA 11.2 version and CUDNN 8.1.0 compatible with corresponding CUDA
- First we install all the dependencies regarding the project onto a virtual env "tensorflow" by
running this on cmd by (assuming anaconda is already installed on your machine)conda create -n tensorflow pip python=3.9
conda activate tensorflow
- Once you have activated your virtual environment,
the name of the environment should be displayed within brackets at the beggining of your cmd path specifier, e.g.:
(tensorflow) F:\Suppi\My_Project>
- Install the Tensorflow pip package by
pip install --ignore-installed --upgrade tensorflow==2.9.0
Note: We need to install Protobuf and Protoc before proceeding with installing other dependencies
* All the Available Info are provided in here TensorFlow 2 Object Detection API tutorial
-
After following through the above processes, just clone this repo and execute Tutorial.ipynb
-
The required checkpoints are already trained and stored in Tensorflow/workspace/models/my_ssd_mobnet/
If you want to build the CNN model yourself try executing Proj.ipynb under the same conda environment (tensorflow)