Road Lane Segmentation with PyTorch

• https://github.com/jaeoh2/Road-Lane-Instance-Segmentation-PyTorch

Road Lane Segmentation with Keras

• Convolution Neural Network for segmentation of the road lane.
• Models : SegNet, U-Net

Load Dataset

def make_dataset(image_list, mask_list, img_size=(360,480)):
    images = []
    masks = []
    
    for img, mask in zip(image_list, mask_list):
         images.append(cv2.resize(cv2.imread(img), img_size))
         masks.append(binarylab(cv2.resize(cv2.imread(mask), img_size)))
        
    images = np.array(images)
    masks = np.array(masks)
    
    return images, masks

Build Model

class_weighting= [0.2595, 0.1826, 4.5640, 0.1417, 0.9051, 0.3826, 9.6446, 1.8418, 0.6823, 6.2478, 7.3614, 1.0974]

model_checkpoint = ModelCheckpoint('model-epoch-{epoch:02d}-val_loss-{val_loss:.2f}.hdf5', monitor='val_loss', save_best_only=True)
model_earlystopping = EarlyStopping(monitor='val_loss')

SegNet

_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
input_2 (InputLayer)         (None, 360, 480, 3)       0         
_________________________________________________________________
conv2d_21 (Conv2D)           (None, 360, 480, 64)      1792      
_________________________________________________________________
batch_normalization_21 (Batc (None, 360, 480, 64)      256       
_________________________________________________________________
activation_21 (Activation)   (None, 360, 480, 64)      0         
_________________________________________________________________
conv2d_22 (Conv2D)           (None, 360, 480, 64)      36928     
_________________________________________________________________
batch_normalization_22 (Batc (None, 360, 480, 64)      256       
_________________________________________________________________
activation_22 (Activation)   (None, 360, 480, 64)      0         
_________________________________________________________________
max_pooling2d_4 (MaxPooling2 (None, 180, 240, 64)      0         
_________________________________________________________________
conv2d_23 (Conv2D)           (None, 180, 240, 128)     73856     
_________________________________________________________________
batch_normalization_23 (Batc (None, 180, 240, 128)     512       
_________________________________________________________________
activation_23 (Activation)   (None, 180, 240, 128)     0         
_________________________________________________________________
conv2d_24 (Conv2D)           (None, 180, 240, 128)     147584    
_________________________________________________________________
batch_normalization_24 (Batc (None, 180, 240, 128)     512       
_________________________________________________________________
activation_24 (Activation)   (None, 180, 240, 128)     0         
_________________________________________________________________
max_pooling2d_5 (MaxPooling2 (None, 90, 120, 128)      0         
_________________________________________________________________
conv2d_25 (Conv2D)           (None, 90, 120, 256)      295168    
_________________________________________________________________
batch_normalization_25 (Batc (None, 90, 120, 256)      1024      
_________________________________________________________________
activation_25 (Activation)   (None, 90, 120, 256)      0         
_________________________________________________________________
conv2d_26 (Conv2D)           (None, 90, 120, 256)      590080    
_________________________________________________________________
batch_normalization_26 (Batc (None, 90, 120, 256)      1024      
_________________________________________________________________
activation_26 (Activation)   (None, 90, 120, 256)      0         
_________________________________________________________________
conv2d_27 (Conv2D)           (None, 90, 120, 256)      590080    
_________________________________________________________________
batch_normalization_27 (Batc (None, 90, 120, 256)      1024      
_________________________________________________________________
activation_27 (Activation)   (None, 90, 120, 256)      0         
_________________________________________________________________
max_pooling2d_6 (MaxPooling2 (None, 45, 60, 256)       0         
_________________________________________________________________
conv2d_28 (Conv2D)           (None, 45, 60, 512)       1180160   
_________________________________________________________________
batch_normalization_28 (Batc (None, 45, 60, 512)       2048      
_________________________________________________________________
activation_28 (Activation)   (None, 45, 60, 512)       0         
_________________________________________________________________
conv2d_29 (Conv2D)           (None, 45, 60, 512)       2359808   
_________________________________________________________________
batch_normalization_29 (Batc (None, 45, 60, 512)       2048      
_________________________________________________________________
activation_29 (Activation)   (None, 45, 60, 512)       0         
_________________________________________________________________
conv2d_30 (Conv2D)           (None, 45, 60, 512)       2359808   
_________________________________________________________________
batch_normalization_30 (Batc (None, 45, 60, 512)       2048      
_________________________________________________________________
activation_30 (Activation)   (None, 45, 60, 512)       0         
_________________________________________________________________
conv2d_31 (Conv2D)           (None, 45, 60, 512)       2359808   
_________________________________________________________________
batch_normalization_31 (Batc (None, 45, 60, 512)       2048      
_________________________________________________________________
activation_31 (Activation)   (None, 45, 60, 512)       0         
_________________________________________________________________
conv2d_32 (Conv2D)           (None, 45, 60, 512)       2359808   
_________________________________________________________________
batch_normalization_32 (Batc (None, 45, 60, 512)       2048      
_________________________________________________________________
activation_32 (Activation)   (None, 45, 60, 512)       0         
_________________________________________________________________
conv2d_33 (Conv2D)           (None, 45, 60, 512)       2359808   
_________________________________________________________________
batch_normalization_33 (Batc (None, 45, 60, 512)       2048      
_________________________________________________________________
activation_33 (Activation)   (None, 45, 60, 512)       0         
_________________________________________________________________
up_sampling2d_4 (UpSampling2 (None, 90, 120, 512)      0         
_________________________________________________________________
conv2d_34 (Conv2D)           (None, 90, 120, 256)      1179904   
_________________________________________________________________
batch_normalization_34 (Batc (None, 90, 120, 256)      1024      
_________________________________________________________________
activation_34 (Activation)   (None, 90, 120, 256)      0         
_________________________________________________________________
conv2d_35 (Conv2D)           (None, 90, 120, 256)      590080    
_________________________________________________________________
batch_normalization_35 (Batc (None, 90, 120, 256)      1024      
_________________________________________________________________
activation_35 (Activation)   (None, 90, 120, 256)      0         
_________________________________________________________________
conv2d_36 (Conv2D)           (None, 90, 120, 256)      590080    
_________________________________________________________________
batch_normalization_36 (Batc (None, 90, 120, 256)      1024      
_________________________________________________________________
activation_36 (Activation)   (None, 90, 120, 256)      0         
_________________________________________________________________
up_sampling2d_5 (UpSampling2 (None, 180, 240, 256)     0         
_________________________________________________________________
conv2d_37 (Conv2D)           (None, 180, 240, 128)     295040    
_________________________________________________________________
batch_normalization_37 (Batc (None, 180, 240, 128)     512       
_________________________________________________________________
activation_37 (Activation)   (None, 180, 240, 128)     0         
_________________________________________________________________
conv2d_38 (Conv2D)           (None, 180, 240, 128)     147584    
_________________________________________________________________
batch_normalization_38 (Batc (None, 180, 240, 128)     512       
_________________________________________________________________
activation_38 (Activation)   (None, 180, 240, 128)     0         
_________________________________________________________________
up_sampling2d_6 (UpSampling2 (None, 360, 480, 128)     0         
_________________________________________________________________
conv2d_39 (Conv2D)           (None, 360, 480, 64)      73792     
_________________________________________________________________
batch_normalization_39 (Batc (None, 360, 480, 64)      256       
_________________________________________________________________
activation_39 (Activation)   (None, 360, 480, 64)      0         
_________________________________________________________________
conv2d_40 (Conv2D)           (None, 360, 480, 64)      36928     
_________________________________________________________________
batch_normalization_40 (Batc (None, 360, 480, 64)      256       
_________________________________________________________________
activation_40 (Activation)   (None, 360, 480, 64)      0         
_________________________________________________________________
output (Conv2D)              (None, 360, 480, 12)      780       
=================================================================
Total params: 17,650,380
Trainable params: 17,639,628
Non-trainable params: 10,752
_________________________________________________________________

result

U-Net

References

• https://github.com/preddy5/segnet/blob/master/Segnet.ipynb
• https://gist.github.com/melgor/0e43cadf742fe3336148ab64dd63138f
• https://github.com/mathildor/TF-SegNet/blob/master/AirNet/layers.py