I recently started going to the gym. I had to rely on more experienced trainers to watch and learn how an equipment is used. I sometimes googled the name written on the body of the equipment to read more details about each equipment, what muscles they trained and videos on how to use them. A thought came to me. Why not build an image recognition web app that recognises an equipment by its photo and outputs details about the equipment? I wrote to a gym owner who let me do the required work such as data gathering, labelling, etc. More details are in Chukwudi_Ajoku_A0151658_Internship_Report.pdf. The above resulted to this project.
This project runs on the browser and requires python.
- Clone the repo
- Create a python environment that meets requirements.txt
- Run init.py
- Go to your browser and open 127.0.0.1:5000/
- Select desired image and submit query.
- Data contains few samples from my train/validation/test sets.
- statis contains the CSS file and cache for uploaded images.
- templates has the HTML templates.
- Chukwudi_Ajoku_A0151658_Internship_Report.pdf is the report.
- Image Classification.ipynb contains the Machine Learning pipeline. From data preparation, training to prediction.
The pretrained model behaves differently when loaded into CPU or GPU. Using TensorFlow with GPU produces desired results as opposed to using TensorFlow on CPU. This is a TensorFlow specific issue and more details can be gotten here.