CWD30

Full Paper arXiv

CWD30 comprises over 219,770 high-resolution images of 20 weed species and 10 crop species, encompassing various growth stages, multiple viewing angles, and environmental conditions. The images were collected from diverse agricultural fields across different geographic locations and seasons, ensuring a representative dataset.

Data Download Link

If you use our data/paper in your projects kindly cite the paper and star the repo.

Global Distribution of Crops in the CWD30 dataset.

MODEL ZOO

Classification Models

⚠️NOTE⚠️ We are currently in middle of uploading the weights. All might not be available.

Model	Weights	Acc
ResNet-18	chkpt	79.5
ResNet-50	chkpt	84.6
ResNet-101	chkpt	81.36
MobileNetv3-S	chkpt	80.5
MobileNetv3-L	chkpt	74.67
EffNet-B0	chkpt	83.2
EffNet-B3	chkpt	83.64
EffNet-B5	chkpt	84.5
ConvNeXt-T	chkpt	85.6
ConvNeXt-M	chkpt	85.9
ConvNeXt-L	chkpt	84.7
ViT-T	chkpt	83.43
ViT-B	chkpt	86.4
CaiT-T	chkpt	85.2
CaiT-S	chkpt	86.9
Swin-T	chkpt	85.59
Swin-B	chkpt	85.3
Swin-L	chkpt	87.0
MaxViT-S	chkpt	86.5
MaxViT-B	chkpt	87.08
CoAtNet-1	chkpt	86.1
CoAtNet-3	chkpt	84.3
EffFormer-L1	chkpt	80.5
EffFormer-L3	chkpt	82.7
EffFormer-L7	chkpt	81.2

Pretrained Weights on iNaturalist

ReNet-101	Weights	Acc.
iNat21	✅chkpt	<80%
iNat17	✅chkpt	60.41%

Semantic Segmentation Models

Access dataset via:

• Sugar Beet
• Carrot Weed
• Bean Weed

⚠️NOTE⚠️ We are currently in middle of uploading the weights. All might not be available.

Model	BeanWeed	SugarBeet	CarrotWeed
UNet	✅72.49 mIOU, chkpt	✅85.47 mIOU, chkpt	✅78.32 mIOU, chkpt
DeepLab v3+	78.03 mIOU, chkpt	86.02 mIOU, chkpt	✅83.16 mIOU, chkpt
OCR	79.51 mIOU, chkpt	87.34 mIOU, chkpt	✅86.53 mIOU, chkpt
SegNext	83.90 mIOU, chkpt	87.65 mIOU, chkpt	✅88.54 mIOU, chkpt

✅MSCAN backbone SegNext

Instances Segmentation Models

Access dataset via:

• PhenoBench
• GrowliFlower

Model	Data	Weights	PQ
MaskRCNN (ResNet-101 FPN backbone)	PhenoBench	✅chkpt	44.05
MaskRCNN (ResNet-101 FPN backbone)	GrowliFlower	✅chkpt	56.33

Citation

@article{ilyas2023cwd30,
  title={CWD30: A Comprehensive and Holistic Dataset for Crop Weed Recognition in Precision Agriculture},
  author={Ilyas, Talha and Arsa, Dewa Made Sri and Ahmad, Khubaib and Jeong, Yong Chae and Won, Okjae and Lee, Jong Hoon and Kim, Hyongsuk},
  journal={arXiv preprint arXiv:2305.10084},
  year={2023}
}

Paper is currently under review. ;)