Camouflaged-Object-Detection

This is a Tensorflow/Keras implementation of the SINet network described in Camouflaged Object Detection

Original implementation in PyTorch may be found in:

Camouflaged Object Detection - CVPR2020-Oral

Link to the database:

COD10k

Link to partially trained network:

Partially trained network ( 500Mb )

Network's Architecture:

Usage:

To train the model:

python3 src/train.py --images dataset/images --labels dataset/objects

Training arguments:

train.py [-h] --images IMAGES --labels LABELS [--epochs EPOCHS] [--batch-size BATCH_SIZE] [--save SAVE] [--check-path CHECK_PATH] [--visualize VISUALIZE] [--save-path SAVE_PATH]

To test the network on a random sample of the validation set after the training use:

--test 1

To retrain a model use:

--retrain 1 --retrain-path 'path_to_the_model'

To visualize the losses after the training use:

--visualize 1

Testing coming soon

Preliminary results: