kirilcvetkov92 / Semantic-Segmentation-BiSeNet

Keras BiseNet architecture implementation

Image-Segmentation

Keras impementation of BiseNet Image Segmentation Model (Paper : Link)

Image	Semantic Segmentation	Semantic_Video_Segmatnation(Youtube)

Pretrained model

Pretrained Model Download (Link)

Dataset

Download CamVid dataset from Semantic-Segmentation-Suite
Thanks GeorgeSeif for his great job!

File hierarchy

Once you download pretrained model and dataset, please follow this project structure:

├── "data"                   
|   ├── train
|   ├── train_labels
|   ├── val
|   ├── val_labels
|   ├── test
|   ├── test_labels
|── "model.h5"   

Support

Prediction supports the following file formats : (Video : Mp4, Picture : .png)

Requirements for loading pretrained model

3 <= Python < 3.6 (Please note that python serialization algorithm is changed from v3.6+, and you can't load the pretrained model if you use Python 3.6+)

Model prediction arguments

mandatory arguments:
  -media MEDIA_DIR, --media_dir MEDIA_DIR
                        Media Directorium for prediction (mp4,png)
optional arguments:
  -save SAVE_DIR, --save_dir SAVE_DIR
                        Save Directorium
  -model MODEL_DIR, --model_dir MODEL_DIR
                        Model Directorium

Example semantic Image segmentation :

python predict.py -media test_img.png

Example semantic Video segmentation :

python predict.py -media test_video.mp4

Model training arguments:

optional arguments:
  -eph EPOCHS, --epochs EPOCHS
                        Number of epochs
  -lr LEARNING_RATE, --learning_rate LEARNING_RATE
                        Learning rate
  -save MODEL_DIR, --model_dir MODEL_DIR
                        Save checkpoints directory
  -batches BATCH_SIZE, --batch_size BATCH_SIZE
                        Number of batches per train

Example training:

python train.py