This repo implement DeepLabv3+ with constrained focal loss in TensorFlow for semantic image segmentation on the anthrax spore dataset. The implementation is largely based on rishizek's DeepLab v3+ implemantation

Please install latest version of TensorFlow and use Python 3.

• For training, you need to download and extract pre-trained Resnet v2 101 model from slim specifying the location with --pre_trained_model.

we use our anthrax spore dataset(https://drive.google.com/open?id=1-Cjy4tkhgBxTip2B_3esqw8xWkzofcZX) to train deeplab v3+ model using constrained focal loss. in anthrax spore dataset: -training set: 400 images -testing set: 200 images There are two classes in each image——background and anthrax spore.

We implement the proposed constrained focal loss in constrained_focal_loss_impl.py.

For training model, you first need to convert original data to the TensorFlow TFRecord format. This enables to accelerate training seep.

python create_pascal_tf_record.py --data_dir DATA_DIR \
                                  --image_data_dir IMAGE_DATA_DIR \
                                  --label_data_dir LABEL_DATA_DIR 

Once you created TFrecord for PASCAL VOC training and validation deta, you can start training model as follow:

python train.py --model_dir MODEL_DIR --pre_trained_model PRE_TRAINED_MODEL

Here, --pre_trained_model contains the pre-trained Resnet model, whereas --model_dir contains the trained DeepLabv3 checkpoints. If --model_dir contains the valid checkpoints, the model is trained from the specified checkpoint in --model_dir.

You can see other options with the following command:

python train.py --help

To evaluate how model perform, one can use the following command:

python evaluate.py --help

To apply semantic segmentation to your images, one can use the following commands:

python inference.py --data_dir DATA_DIR --infer_data_list INFER_DATA_LIST --model_dir MODEL_DIR 

Pull requests are welcome.

• Freeze batch normalization during training
• Multi-GPU support
• Channels first support (Apparently large performance boost on GPU)
• Model pretrained on MS-COCO
• Unit test

This repo borrows code heavily from

• rishizek's DeepLab v3+ implemantation

first commit