README.md

metadata

license: cc-by-nc-sa-4.0
task_categories:
  - image-segmentation
  - image-classification
tags:
  - weed
  - agriculture
  - segmentation
  - semantic-segmentation
  - height-estimation
  - regression
  - growth-stage
  - multi-task-learning
  - weedsense
  - precision-agriculture
  - plant-phenotyping
  - temporal
pretty_name: 'WeedSense: Multi-Task Weed Analysis Dataset'
size_categories:
  - 100K<n<1M

WeedSense Dataset

Multi-task temporal dataset of 16 weed species for semantic segmentation, height regression, and growth stage classification from the WeedSense paper:

WeedSense: Multi-Task Learning for Weed Segmentation, Height Estimation, and Growth Stage Classification Toqi Tahamid Sarker, Khaled R Ahmed, Taminul Islam, Cristiana Bernardi Rankrape, Karla Gage Southern Illinois University Carbondale, USA ICCV 2025

Paper (arXiv) | Project Page | Code

Overview

Property	Value
Total frames	120,341
Weed species	16
Growth duration	11 weeks
Total videos	349
Frame resolution	720 x 960 pixels
Annotation types	Segmentation masks, Height (cm), Growth stage (week)
Semantic classes	17 (16 species + background)
Height range	0.2 - 155 cm

Data Splits

Split	Images	Percentage
Train	96,134	~80%
Validation	12,333	~10%
Test	11,874	~10%

Folder Structure

Large folders are distributed as zip archives (HF enforces a 10,000-file-per-directory limit).

train/
  images.zip              # RGB images (720 x 960) as .jpg
  masks.zip               # Segmentation masks (single-channel, class IDs 0-16) as .png
  mmseg_masks.zip         # Segmentation masks formatted for MMSegmentation as .png
  xml.zip                 # VOC-format bounding box annotations as .xml
  train_data.csv          # Metadata: img, species, height, week
val/
  images.zip
  masks.zip
  mmseg_masks.zip
  xml.zip
  val_data.csv
test/
  images.zip
  masks.zip
  mmseg_masks.zip
  xml.zip
  test_data.csv

After Extracting

Each zip extracts into a flat list of files. For example, train/images.zip contains:

ABUTH_week_10_IMG_1656_frame_0.jpg
ABUTH_week_10_IMG_1656_frame_1.jpg
SETFA_week_8_IMG_1344_frame_13.jpg
...

File Naming Convention

All files follow the pattern: {SPECIES}_week_{WEEK}_IMG_{VIDEO_ID}_frame_{FRAME}.{ext}

Example: SETFA_week_8_IMG_1344_frame_13.jpg

CSV Metadata

Each split CSV contains columns:

Column	Type	Description
img	string	Image filename (e.g., ABUTH_week_10_IMG_1656_frame_0.jpg)
species	string	Weed species EPPO code (e.g., ABUTH, SETFA)
height	float	Plant height in centimeters (0.2 - 155.0 cm)
week	int	Growth stage week (1 - 11)

Segmentation Mask Values

Pixel Value	Class
0	Background
1	ABUTH (Velvetleaf)
2	AMAPA (Palmer Amaranth)
3	AMARE (Redroot Pigweed)
4	AMATA / AMATU (Tall Waterhemp)
5	AMBEL (Common Ragweed)
6	CHEAL (Common Lambsquarters)
7	CYPES (Yellow Nutsedge)
8	DIGSA (Large Crabgrass)
9	ECHCG (Barnyardgrass)
10	ERICA (Horseweed)
11	PANDI (Fall Panicum)
12	SETFA (Giant Foxtail)
13	SETPU (Yellow Foxtail)
14	SIDSP (Prickly Sida)
15	SORHA (Johnsongrass)
16	SORVU (Shattercane)

Weed Species Summary

EPPO Code	Scientific Name	Max Height (cm)	Growth Rate (cm/week)	Category
AMATU	Amaranthus tuberculatus	155.0	13.72	Fast
SORHA	Sorghum halepense	121.0	14.06	Fast
SETFA	Setaria faberi	124.0	11.75	Fast
SORVU	Sorghum bicolor	100.0	9.84	Medium
PANDI	Panicum dichotomiflorum	87.0	8.40	Medium
SETPU	Setaria pumila	99.0	8.20	Medium
DIGSA	Digitaria sanguinalis	77.0	7.53	Medium
ECHCG	Echinochloa crus-galli	80.0	7.38	Medium
SIDSP	Sida spinosa	69.0	6.77	Medium
AMARE	Amaranthus retroflexus	75.0	6.86	Medium
ABUTH	Abutilon theophrasti	72.0	6.32	Medium
AMBEL	Ambrosia artemisiifolia	71.0	6.19	Medium
AMAPA	Amaranthus palmeri	62.0	5.66	Slow
CYPES	Cyperus esculentus	56.0	5.42	Slow
CHEAL	Chenopodium album	30.0	2.86	Slow
ERICA	Erigeron canadensis	17.3	1.70	Slow

Data Collection

• Location: SIU Horticulture Research Center greenhouse, Southern Illinois University Carbondale, USA
• Equipment: iPhone 15 Pro Max positioned 1.5 feet above specimens
• Capture: 360-degree video at 1440 x 1920 resolution, 30 FPS
• Environment: 1000W HPS grow lights, 30-32 degree C
• Preprocessing: Temporal downsampling (every 2nd frame), spatial downscaling to 720 x 960
• Annotation: SAM2-Hiera-L semi-automatic segmentation with manual verification and correction
• Height: 325 manual weekly measurements (0.2 - 155 cm)

Benchmark Results (WeedSense Model)

Task	Metric	Value
Segmentation	mIoU	89.78%
Segmentation	mF1	94.54%
Height Estimation	MAE	1.67 cm
Height Estimation	RMSE	2.32 cm
Height Estimation	R squared	0.9941
Growth Stage	Accuracy	99.99%
Growth Stage	F1	99.99%

Usage

from huggingface_hub import snapshot_download, hf_hub_download
import zipfile, os

# Download entire dataset
local_dir = snapshot_download(repo_id="baselab/weedsense", repo_type="dataset")

# Extract a zip file
with zipfile.ZipFile(os.path.join(local_dir, "train", "images.zip"), "r") as z:
    z.extractall(os.path.join(local_dir, "train", "images"))

# Load metadata
import pandas as pd
train_csv = hf_hub_download(
    repo_id="baselab/weedsense",
    repo_type="dataset",
    filename="train/train_data.csv",
)
df = pd.read_csv(train_csv)
print(df.head())
#                                        img species  height  week
# 0  ABUTH_week_10_IMG_1656_frame_0.jpg   ABUTH    50.0    10
# 1  ABUTH_week_10_IMG_1656_frame_1.jpg   ABUTH    50.0    10

Citation

If you use this dataset, please cite:

@inproceedings{sarker2025weedsense,
  title={Weedsense: Multi-task learning for weed segmentation, height estimation, and growth stage classification},
  author={Sarker, Toqi Tahamid and Ahmed, Khaled R and Islam, Taminul and Rankrape, Cristiana Bernardi and Gage, Karla},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={7180--7190},
  year={2025}
}

License

This dataset is released under the CC BY-NC-SA 4.0 license. Commercial use is not permitted. Derivative works must use the same license.