Converting MVCNN from pytorch to Tensorflow 2.11. Transformation is based on this repo.
Dataset structure should be as following:
./
└── data/
├── train/
│ └── class/
│ └── mv1/
│ ├── img1.jpg
│ ├── img2.jpg
│ └── imgn.jpg
└── test/
└── class/
└── mv1/
├── img1.jpg
├── img2.jpg
└── imgn.jpg
Train
python train.py --train_dir path/to/train/dataset --model mobilenet --epochs 10
Predict
python predict.py --model mobilenet --model_weights path/to/model/weights --mvs path/to/image/folder
evaluate.py
script haven not been tested. Try at your own risk.
Dockerfile follows SageMaker requirements. This is to allow TrainingJob can be called from SageMaker estimator.