Deep learning project about animal behaviour.
For now, the project includes:
- EfficientDet model (no backbone) training and saving
- EfficientDet model (no backbone) evaluation (including metrics and image testing)
- EfficientNet backbone training and saving
- EfficientDet with EfficientNet backbone training and saving
- EfficientDet with EfficientNet backbone evaluation (including metrics and image testing)
Before you begin, ensure you have met the following requirement:
- Python 3.10+
Step-by-step guide on how to install the project:
git clone https://github.com/antoinedenovembre/NTNU_project.git
cd NTNU_project
python install_requirements.py/!\ Important: You also need to make sure you have the following structure, including the data folder
NTNU_project_light
│ README.md
│ main.py
│ install_requirements.py
│
└───data
│ └───backbone
│ │ │ image1.jpg/png/...
│ │ │ ...
│ │
│ └───effdet
│ └─── train
│ │ └─── annotations
│ │ │ │ annotations.json
│ │ │ │ ...
│ │ │
│ │ └─── images
│ │ │ image1.jpg/png/...
│ │ │ ...
│ │
│ └─── test
│ └─── annotations
│ │ │ annotations.json
│ │ │ ...
│ │
│ └─── images
│ │ image1.jpg/png/...
│ │ ...
│
└───efficient_det
│
└───barlow
│
└───utils
│
└───scripts
│
└───documentation
│
└───output
The annotations shall have the following structure
{
"annotations": [
{
"area": 87292,
"bbox": [
576,
98,
547,
204
],
"category_id": 6,
"id": 1, <!-- Should be the number of the annotation -->
"image_id": 2, <!-- Should be the number of the image -->
"iscrowd": 0
},
...
]
}
Here is the command to run the project:
python main.py
Overall view of the project: Documentation - Overall view
Technical documentation about evaluation metrics and loss function: Documentation - Metrics
- Ideally one with a sufficient GPU, like NVIDIA RTX 2080Ti
- Any Linux distro should do the trick, but I recommend using Ubuntu 22.04+
This repository is a fork from this repository
Show your appreciation to those who have contributed to the project.
- @FayazRahman - Contribution
- Dev Narayan Chakkalakkal Binu - Contribution
- @antoinedenovembre - Contribution