VegAnn

Vegetation Annotation of a large multi-crop RGB Dataset acquired under diverse conditions for image semantic segmentation

Table of contents

1. Keypoints
2. Abstract
3. Pytorch Data Loader
4. Baseline Results
5. Citing
6. Paper
7. Meta-Information

⏳ Keypoints

• The dataset can be accessed at https://doi.org/10.5281/zenodo.7636408.
• VegAnn contains 3775 images
• Images are 512*512 pixels
• Corresponding binary masks is 0 for soil + crop residues (background) 255 for Vegetation (foreground)
• The dataset includes images of 26+ crop species, which are not evenly represented
• VegAnn was compiled using a variety of outdoor images captured with different acquisition systems and configurations
• For more information about VegAnn, details, labeling rules and potential uses see @paper

📚 Abstract

Applying deep learning to images of cropping systems provides new knowledge and insights in research and commercial applications. Semantic segmentation or pixel-wise classification, of RGB images acquired at the ground level, into vegetation and background is a critical step in the estimation of several canopy traits. Current state of the art methodologies based on convolutional neural networks (CNNs) are trained on datasets acquired under controlled or indoor environments. These models are unable to generalize to real-world images and hence need to be fine-tuned using new labelled datasets. This motivated the creation of the VegAnn - Vegetation Annotation - dataset, a collection of 3795 multi-crop RGB images acquired for different phenological stages using different systems and platforms in diverse illumination conditions. We anticipate that VegAnn will help improving segmentation algorithm performances, facilitate benchmarking and promote large-scale crop vegetation segmentation research.

📦 Pytorch Data Loader

We provide Python dataloader that load the data as PyTorch tensors. With the dataloader, users can select desired images with the metadata information such as species, camera system, and training/validation/test sets.

🍲 Example use :

Here is an example use case of the dataloader with our custom dataset class:

    from segmentation_models_pytorch.encoders import get_preprocessing_fn
    from utils.dataset import DatasetVegAnn
    from torch.utils.data import DataLoader

    train_dataset = DatasetVegAnn(images_dir = veganpath,species = ["Wheat","Maize"], system = ["Handeld Cameras","Phone Camera"], tvt="Training")    
    train_dataloader = DataLoader(train_dataset, batch_size=16, shuffle=True,pin_memory=False, num_workers=10)

By using this dataloader, you can easily load the desired images as PyTorch tensors see utils/dataset.py for more details.

👀 Baseline Results

Metrics are computed at the dataset level for the 5 Test sets of VegAnn

Method	Encoder	IOU	F1
Unet	ResNet34	89.7 ±1.4	94.5 ±0.8
DeepLabV3	ResNet34	89.5 ±0.2	94.5 ±0.2

📝 Citing

If you find this dataset useful, please cite:

@article{madec2023, title={VegAnn: Vegetation Annotation of multi-crop RGB images acquired under diverse conditions for segmentation}, author={Madec, Simon and Irfan, Kamran and Velumani, Kaaviya and Baret, Frederic and David, Etienne and Daubige, Gaetan and Samatan, Lucas and Serouart, Mario and Smith, Daniel and James, Chris and Camacho, Fernando and Guo, Wei and De Solan, Benoit and Chapman, Scott and Weiss, Marie }, url={https://doi.org/10.5281/zenodo.7636408}, year={2023} }

📖 Paper

In review stage

☸️ Model inference

Docker image in construction