Upload 12 files

Benchmark/Classification-Wheat Disease/readme.txt

@@ -0,0 +1,8 @@
+Wheat Disease Classification. 
+The second task is wheat disease classification. We collected images from five public datasets of CerealConv, WFD, AWDD, MSWDD, and LWDCD, to construct a comprehensive dataset of healthy wheat and other eight disease types including Brownrust, Mildew, Septoria, Yellowrust, Stemrust, Healthy, Wheatscab, and Yellowdwarf, with 4000 images in total. These images were obtained from multiple countries in Europe, North America, and Asia between 2019 and 2024, reflecting the sensitivity and resistance of different wheat varieties to diseases and covering images of different varieties, periods, and environmental conditions. Based on this dataset, we established the Wheat Disease Classification benchmark, partitioned into 320 training samples and 3,680 testing samples. The identical random five-fold sampling trials were conducted to ensure statistical soundness.
+
+Long, M., Hartley, M., Morris, R. J. & Brown, J. K. M. Classification of wheat diseases using deep learning networks with field and glasshouse images. Plant Pathol. 72, 536–547 (2023).
+Genaev, M. A. et al. Image-based wheat fungi diseases identification by deep learning. Plants 10, 1500 (2021).
+Safarijalal, B., Alborzi, Y. & Najafi, E. Automated Wheat Disease Detection using a ROS-based Autonomous Guided UAV. Preprint at http://arxiv.org/abs/2206.15042 (2022).
+Mao, R. et al. DAE-Mask: a novel deep-learning-based automatic detection model for in-field wheat diseases. Precis. Agric. 25, 785–810 (2024).
+Goyal, L., Sharma, C. M., Singh, A. & Singh, P. K. Leaf and spike wheat disease detection & classification using an improved deep convolutional architecture. Inform. Med. Unlocked 25, 100642 (2021).

Benchmark/Classification-Wheat Growth Stage/readme.txt

@@ -0,0 +1,5 @@
+Wheat Growth Stage Classification. 
+The first task is wheat growth stage detection where the model is tasked to predict the correct growth stage. We adopt the WGSP dataset for wheat growth stage classification. The dataset contains 3353 RGB images of wheat canopies collected from field trials conducted in China, the United Kingdom, and the United States between 2018 and 2021. Spanning four growing seasons, the dataset covers 263 wheat varieties and encompasses six critical growth stages: tillering, jointing, booting, Anthesis, grain filling, and maturity. The dataset reflects a wide range of geographic locations, climatic conditions, planting densities, and canopy color and structural characteristics across different wheat varieties. We partition it into 671 training images and 2,682 testing images. Under a reduced training data protocol, that is, 75%, 50%, and 25% data are sampled from the original training set, we performed five independent random sampling per data fraction to mitigate stochastic sampling bias.
+
+
+Shen, L. et al. GSP-AI: An AI-Powered Platform for Identifying Key Growth Stages and the Vegetative-to-Reproductive Transition in Wheat Using Trilateral Drone Imagery and Meteorological Data. Plant Phenomics 6, 0255 (2024).

Benchmark/Counting-Rice Leaf/Rice Leaf.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6a7ce39feeaca83b8c15116d59eb04437269f183fb16770c78c06178d800ed24
+size 1255323080

Benchmark/Counting-Rice Leaf/readme.txt

@@ -0,0 +1,2 @@
+Rice Leaf Counting. 
+To justify cross-species generalization of FoMo4Wheat, we also evaluate the counting performance on rice crop. Between 2020 and 2022, our team conducted field experiments in Nanjing, Sanya, Danyang, covering over 1,000 rice varieties. Using drones, we collected canopy structure data at different growth stages of rice prior to tillering. We call this dataset as the Rice Leaf Counting dataset, partitioned into 1,360 training images and 340 testing images.

Benchmark/Counting-Wheat Leaf/readme.txt

@@ -0,0 +1,4 @@
+Wheat Leaf Counting. 
+We first evaluate on wheat leaf counting where a model is required to count all leaf tips. Between 2020 and 2022, our team conducted field experiments in Xuzhou, Baima, Henan, and Jurong, China, and created a leaf counting benchmark using our self-developed PhenoArm imaging equipment and outdoor monitoring devices with the Hunting camera. Image data on the canopy structure at different growth stages before wheat tillering were captured from 0° and 45° angles. Each leaf tip is labeled with a dot. Five experimenters were involved in data labelling and were cross-checked with each other for correctness. 1,508 images were used for training, and 379 images are used for testing.
+
+Li, Y. et al. Self-Supervised Plant Phenotyping by Combining Domain Adaptation with 3D Plant Model Simulations: Application to Wheat Leaf Counting at Seedling Stage. Plant Phenomics 5, 0041 (2023).

Benchmark/Detection-Aerial Wheat Spikes/Aerial Wheat Spikes.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4f0dea99bebb302f1a5036c7495cb8da59a240fa9697288f805998d7cf28d8d9
+size 79191624

Benchmark/Detection-Aerial Wheat Spikes/readme.txt

@@ -0,0 +1,3 @@
+Wheat Spike Detection From UAV Imagery. 
+From 2021 to 2024, our team conducted field experiments in various locations including Yangling, Nanchong, Baima, and Jurong, China, and developed a drone-based wheat spike detection dataset. Visible light images were captured at heights of both six and ten meters, with a resolution of 4608x3456 pixels. The high-resolution images were then cropped to 224x224 pixels to highlight wheat spikes. To assess the cross-domain generalization of ground-based wheat head detection models, we established a UAV-based testing set comprising images acquired at different altitudes: 100 images at GSD 0.6mm elevation (Nanchong) and 120 images at GSD 1.2mm elevation (Yangling).
+

Benchmark/Detection-Ground-based Wheat Spikes/readme.txt

@@ -0,0 +1,4 @@
+Wheat Head Detection From Ground-based Imagery. 
+We use the Global Wheat Head Detection (GWHD)  Dataset to evaluate wheat head detection. The dataset is developed for an international competition, with 4,700 high-resolution RGB images from various locations around the world between 2016 and 2019, containing 190,000 labeled wheat heads. The images cover multiple genotypes at different growth stages and include data from several countries, including Europe, North America, Australia, and Asia, under various planting densities and environmental conditions. The dataset is divided into 3,607 training images, 1,448 validation images, and 1,382 testing images.
+
+David, E. et al. Global Wheat Head Detection 2021: An Improved Dataset for Benchmarking Wheat Head Detection Methods. Plant Phenomics 2021, 2021/9846158 (2021).

Benchmark/Segmentation-Crop and Weed/readme.txt

@@ -0,0 +1,4 @@
+Crop and Weed Segmentation. 
+The Crop and Weed dataset is adopted, which is a large-scale, multi-modal dataset designed for crop and weed segmentation. The data was collected from multiple commercial agricultural sites and specially cultivated in outdoor experimental plots in Austria, spanning various growing seasons between 2018 and 2021. It covers diverse lighting conditions, weather scenarios, soil types, and combinations of crops and weeds, while also including negative samples with no visible vegetation and backgrounds containing debris. The dataset contains 7,705 images and approximately 112,000 annotated instances. We adopt the Fine24 subvariant of the Crop and Weed dataset as the benchmark (16 crop and 8 weed categories), partitioned into 6,164 training images and 1,541 test images.
+
+Steininger, D., Trondl, A., Croonen, G., Simon, J. & Widhalm, V. The CropAndWeed Dataset: a Multi-Modal Learning Approach for Efficient Crop and Weed Manipulation. in 2023 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) 3718–3727 (IEEE, Waikoloa, HI, USA, 2023). doi:10.1109/WACV56688.2023.00372.

Benchmark/Segmentation-Muiti-Crop/readme.txt

@@ -0,0 +1,4 @@
+Multi-Crop Segmentation. 
+We use the VegAnn dataset, which is a binary segmentation dataset designed to differentiate vegetation, including both healthy and senescent plant parts, from background elements such as soil and crop residues. It comprises 3,775 high-resolution RGB images collected from diverse regions using various acquisition systems and platforms. The dataset encompasses 9 crop types, spanning different growth stages, climatic conditions, and soil types, and includes data from multiple countries such as France, China, Japan, Belgium, Australia, and others. The dataset is partitioned into 3,020 training images and 755 test images.
+
+Madec, S. et al. VegAnn, Vegetation Annotation of multi-crop RGB images acquired under diverse conditions for segmentation. Sci. Data 10, 302 (2023).

Benchmark/Segmentation-Rice Organ/readme.txt

@@ -0,0 +1,4 @@
+Rice Organ Segmentation. 
+We validate on another crop variety to justify the cross-species generalization of the backbone. Our team developed the RiceSEG image segmentation dataset through international collaboration, focusing on rice fields with weeds and duckweed backgrounds. The images were collected between 2012 and 2023 by 10 institutions from 10 locations across five countries, including China, Japan, India, the Philippines, and Tanzania, covering over 1,000 rice varieties. The images were captured using various camera types, such as digital SLRs, portable action cameras, and smartphones. The cameras were positioned 1-2 meters above the canopy and oriented toward the canopy in different directions (0°-90°). The dataset has 2,462 training images and 616 testing images. 
+
+Zhou, J. et al. Global Rice Multiclass Segmentation Dataset (RiceSEG): Comprehensive and Diverse High-Resolution RGB-Annotated Images for the Development and Benchmarking of Rice Segmentation Algorithms. Plant Phenomics 100099 (2025) doi:10.1016/j.plaphe.2025.100099.

Benchmark/Segmentation-Wheat Organ/readme.txt

@@ -0,0 +1,4 @@
+Wheat Organ Segmentation. 
+The Global Wheat Full Semantic Segmentation (GWFSS) dataset is employed. It is a recently released international competition dataset that includes 1,096 high-resolution RGB images collected from around the world between 2020 and 2024. The dataset contains four categories for wheat organ segmentation. The images cover a variety of genotypes at different growth stages and include data from collaborative institutions across Switzerland, Belgium, UK, China , Mexico, Australia, France, Morocco, Japan, Canada collected under various planting densities and environmental conditions. The dataset is partitioned into 876 training images and 220 testing images.
+
+Wang, Z. et al. The Global Wheat Full Semantic Organ Segmentation (GWFSS) Dataset. 2025.03.18.642594 Preprint at https://doi.org/10.1101/2025.03.18.642594 (2025).