ObjectTag makes it easy to create high-quality datasets for object detection. Capture images, annotate objects with bounding boxes, add labels, review your data, and export your dataset in common formats like YOLO, COCO or as needed for PyTorch.
Whether you are prototyping a computer vision model, collecting training data in the field, or preparing a dataset for ML workflows, ObjectTag keeps the process simple and efficient.
Download iOS app from AppStore
- Download and unzip your dataset,
- Copy ObjectDetectionDataset or create your own dataset class
- Use it e.g. like this in PyTorch:
from ODDataset import ObjectDetectionDataset
from torch.utils.data import DataLoader
from torchvision import transforms
default_transform = transforms.Compose([
transforms.ToTensor(),
])
def collate_fn(batch):
return tuple(zip(*batch))
# Expected structure:
# mydataset/
# ├── dimensions/
# │ ├── image_001.jpg
# │ └── image_002.jpg
# └── labels.json
dataset = ObjectDetectionDataset(
image_folder="mydataset/dimensions",
labels_json_path="mydataset/labels.json",
transform=default_transform,
)
loader = DataLoader(
dataset,
batch_size=4,
shuffle=True,
collate_fn=collate_fn,
)
for images, targets in loader:
print(f"Images batch shape: {images[0].shape}")
print(f"Targets: {targets, }")
break # just the first batch for testingIf you have feature requests or encounter bugs, please open an issue in this repository.