CamVid dataset: Example of the aitlas toolbox for semantic segmentation

This notebook shows a sample implementation of a image segmentation using the aitlas toolbox.

Import the required packages

from aitlas.datasets import CamVidDataset
from aitlas.models import DeepLabV3
from aitlas.transforms import MinMaxNormTranspose
from aitlas.utils import image_loader

Visualize images and masks

dataset_config = {
    "data_dir": "../data/camvid/train"
}
dataset = CamVidDataset(dataset_config)

print(f"Total number of patches: {len(dataset)}")
dataset.show_image(10);
dataset.show_image(26);

Total number of patches: 367

dataset.data_distribution_table()

	Number of pixels
sky	10682767.0
building	14750079.0
column_pole	623349.0
road	20076880.0
sidewalk	2845085.0
tree	6166762.0
sign	743859.0
fence	714595.0
car	3719877.0
pedestrian	405385.0
byciclist	184967.0
void	2503995.0

dataset.data_distribution_barchart();

Load train data

train_dataset_config = {
    "batch_size": 16,
    "shuffle": True,
    "data_dir": "../data/camvid/train",
    "transforms": ["aitlas.transforms.MinMaxNormTranspose"],
    "target_transforms": ["aitlas.transforms.Transpose"]
}
train_dataset = CamVidDataset(train_dataset_config)

validation_dataset_config = {
    "batch_size": 16,
    "shuffle": False,
    "data_dir": "../data/camvid/val",
    "transforms": ["aitlas.transforms.MinMaxNormTranspose"],
    "target_transforms": ["aitlas.transforms.Transpose"]
}
validation_dataset = CamVidDataset(validation_dataset_config)

len(train_dataset), len(validation_dataset)

(367, 101)

Create the model

epochs = 100
model_directory = "./experiments/camvid"
model_config = {
    "num_classes": 12,
    "learning_rate": 0.0001,
    "pretrained": True,
    "threshold": 0.5,
    "metrics": ["iou"]
}

model = DeepLabV3(model_config)
model.prepare()

Start the training

model.train_and_evaluate_model(
    train_dataset=train_dataset,
    val_dataset=validation_dataset,
    epochs=epochs,
    model_directory=model_directory,
    run_id='1'
);

Evalute the model using test data

test_dataset_config = {
    "batch_size": 4,
    "shuffle": False,
    "num_workers": 4,
    "data_dir": "../data/camvid/test",
    "transforms": ["aitlas.transforms.MinMaxNormTranspose"],
    "target_transforms": ["aitlas.transforms.Transpose"]
}

test_dataset = CamVidDataset(test_dataset_config)
len(test_dataset)

model = DeepLabV3(model_config)
model.prepare()
model.running_metrics.reset()
model_path = "./experiments/camvid/checkpoint.pth.tar"
model.evaluate(dataset=test_dataset, model_path=model_path)
model.running_metrics.get_scores(model.metrics)

Predictions

model_path = "./experiments/camvid/checkpoint.pth.tar"
labels = CamVidDataset.labels
transform = MinMaxNormTranspose()
model.load_model(model_path)

image = image_loader('../data/camvid/test/images/Seq05VD_f02520.png')
fig = model.predict_masks(image, labels, transform)

image = image_loader('../data/camvid/test/images/0001TP_009450.png')
fig = model.predict_masks(image, labels, transform)

image = image_loader('../data/camvid/test/images/Seq05VD_f01500.png')
fig = model.predict_masks(image, labels, transform)

image = image_loader('../data/camvid/test/images/Seq05VD_f00180.png')
fig = model.predict_masks(image, labels, transform)

2022-10-31 16:52:34,140 INFO Loading checkpoint ./experiments/camvid/checkpoint.pth.tar
2022-10-31 16:52:34,663 INFO Loaded checkpoint ./experiments/camvid/checkpoint.pth.tar at epoch 101