Weedsense Eval

weedsense-eval

WeedSense: Multi-Task Learning for Weed Segmentation, Height Estimation, and Growth Stage Classification — Sarker et al. (2025) (arXiv:2508.14486, 2025)

What this evaluates

Evaluates a model's ability to jointly perform fine-grained weed species segmentation, continuous plant height regression, and discrete temporal growth stage classification from single RGB images.

Datasets

• WeedSense — total ?; splits: test (-1)

Metrics

• mIoU (primary) — range: percent
  • Mean Intersection over Union averaged across all 17 classes (including background). Computed as the average of IoU per class over the dataset.
• MAE (primary) — range: other
  • Mean Absolute Error between predicted and ground-truth plant height in centimeters.
• Accuracy (primary) — range: percent
  • Percentage of correctly predicted growth stage labels (weekly intervals 1–11) out of total samples.

Input / output format

Input: Single RGB image of a weed plant in a greenhouse setting.

Output: Per image: (1) pixel-wise segmentation mask with 17 classes, (2) scalar plant height in cm, (3) discrete growth stage class label (weeks 1–11).

Scoring recipe

ious = []
for c in range(17):
    p = (pred_mask == c)
    g = (gold_mask == c)
    inter = np.sum(p & g)
    union = np.sum(p | g)
    ious.append(inter / union if union > 0 else 1.0)
mIoU = np.mean(ious) * 100
MAE = np.mean(np.abs(pred_height - gold_height))
Accuracy = np.mean(pred_stage == gold_stage) * 100

Common pitfalls

• The dataset includes 16 weed species plus a background class, making it a 17-class segmentation problem; failing to account for background inflates mIoU.
• Growth stage labels are discrete weekly intervals (1–11) rather than continuous time; treating them as regression targets will yield incorrect Accuracy/F1 scores.
• Height is measured in centimeters and reported as MAE; confusing it with RMSE or R² will misrepresent the primary regression metric.

Evidence (verbatim from paper)

mIoU(%)↑ mF1(%)↑ MAE(cm)↓ RMSE(cm)↓ R2↑ Max Error(cm)↓ Within1cm (%)↑ Within2cm (%)↑ Within5cm (%)↑ Accuracy(%)↑ F1(%)↑

Citation

@misc{sarker2025weedsense,
  title={WeedSense: Multi-Task Learning for Weed Segmentation, Height Estimation, and Growth Stage Classification},
  author={Sarker et al. (2025)},
  year={2025},
  note={arXiv:2508.14486}
}

• arXiv: 2508.14486