Example of the aitlas toolbox in the context of object detection in remote sensing images

Import the required packages

import warnings
warnings.filterwarnings("ignore", category=UserWarning)
warnings.filterwarnings("ignore", category=DeprecationWarning)

from aitlas.datasets import ObjectDetectionPascalDataset
from aitlas.transforms import Resize
from aitlas.utils import image_loader
from aitlas.models import FasterRCNN

Visualize images and objects witn bounding boxes

dataset_config = {
    "image_dir": "../data/HRRSD/JPEGImages/",
    "annotations_dir": "../data/HRRSD/TGRS-HRRSD-Dataset/OPT2017/Annotations/",
    "imageset_file": "../data/HRRSD/TGRS-HRRSD-Dataset/OPT2017/ImageSets/Main/test.txt"
}
dataset = ObjectDetectionPascalDataset(dataset_config)

print(f"Total number of patches: {len(dataset)}")
dataset.show_image(5);
dataset.show_image(800);
dataset.show_image(2458);

Total number of patches: 10943

dataset.show_batch(15);

dataset.data_distribution_table()

	Label	Count
0	T junction	2475
1	airplane	2452
2	baseball diamond	2032
3	basketball court	2249
4	bridge	2408
5	crossroad	2659
6	ground track field	2281
7	harbor	1985
8	parking lot	2480
9	ship	2078
10	storage tank	2232
11	tennis court	2322
12	vehicle	2390

dataset.data_distribution_barchart();

Load train data

train_dataset_config = {
    "batch_size": 16,
    "shuffle": True,
    "image_dir": "../data/HRRSD/JPEGImages/",
    "annotations_dir": "../data/HRRSD/TGRS-HRRSD-Dataset/OPT2017/Annotations/",
    "imageset_file": "../data/HRRSD/TGRS-HRRSD-Dataset/OPT2017/ImageSets/Main/train.txt",
    "joint_transforms": ["aitlas.transforms.ResizeToTensorV2"]
}
train_dataset = ObjectDetectionPascalDataset(train_dataset_config)

validation_dataset_config = {
    "batch_size": 16,
    "shuffle": True,
    "image_dir": "../data/HRRSD/JPEGImages/",
    "annotations_dir": "../data/HRRSD/TGRS-HRRSD-Dataset/OPT2017/Annotations/",
    "imageset_file": "../data/HRRSD/TGRS-HRRSD-Dataset/OPT2017/ImageSets/Main/val.txt",
    "joint_transforms": ["aitlas.transforms.ResizeToTensorV2"]
}
validation_dataset = ObjectDetectionPascalDataset(validation_dataset_config)

print(f"Total number of train images: {len(train_dataset)}, Total number of validation images: {len(validation_dataset)}")

Total number of train images: 5401, Total number of validation images: 5417

Create the model

epochs = 20
model_directory = "./experiments/hrrsd"
model_config = {
    "num_classes": 14,
    "learning_rate": 0.0001,
    "pretrained": True,
    "metrics": ["map"]
}

model = FasterRCNN(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 = {
    "shuffle": False,
    "image_dir": "../data/HRRSD/JPEGImages/",
    "annotations_dir": "../data/HRRSD/TGRS-HRRSD-Dataset/OPT2017/Annotations/",
    "imageset_file": "../data/HRRSD/TGRS-HRRSD-Dataset/OPT2017/ImageSets/Main/test.txt",
    "joint_transforms": ["aitlas.transforms.ResizeToTensorV2"]

}

test_dataset = ObjectDetectionPascalDataset(test_dataset_config)
print(f"Total number of test images: {len(test_dataset)}")

model = FasterRCNN(model_config)
model.prepare()
model.running_metrics.reset()
model_path = "./experiments/hrrsd-10-4/best_checkpoint_1670264252_17.pth.tar"
model.evaluate(dataset=test_dataset, model_path=model_path)
model.running_metrics.get_scores(model.metrics)

Predictions

model_config = {
    "num_classes": 14,
    "learning_rate": 0.0001,
    "pretrained": True,
    "metrics": ["map"]
}
model = FasterRCNN(model_config)
model.prepare()

model_path = "./experiments/hrrsd-10-4/best_checkpoint_1670264252_17.pth.tar"
labels = [None,
         'T junction',
         'airplane',
         'baseball diamond',
         'basketball court',
         'bridge',
         'crossroad',
         'ground track field',
         'harbor',
         'parking lot',
         'ship',
         'storage tank',
         'tennis court',
         'vehicle']
transform = Resize()
model.load_model(model_path)

image = image_loader('../data/HRRSD/JPEGImages/00042.jpg')
fig = model.detect_objects(image, labels, transform)

image = image_loader('../data/HRRSD/JPEGImages/00620.jpg')
fig = model.detect_objects(image, labels, transform)

image = image_loader('../data/HRRSD/JPEGImages/19739.jpg')
fig = model.detect_objects(image, labels, transform)

image = image_loader('../data/HRRSD/JPEGImages/20979.jpg')
fig = model.detect_objects(image, labels, transform)

image = image_loader('../data/HRRSD/JPEGImages/10063.jpg')
fig = model.detect_objects(image, labels, transform)

image = image_loader('../data/HRRSD/JPEGImages/09536.jpg')
fig = model.detect_objects(image, labels, transform)

2022-12-07 12:18:11,535 INFO Loading checkpoint ./experiments/hrrsd-10-4/best_checkpoint_1670264252_17.pth.tar
2022-12-07 12:18:11,832 INFO Loaded checkpoint ./experiments/hrrsd-10-4/best_checkpoint_1670264252_17.pth.tar at epoch 18