# Beyond Visible Spectrum: AI for Agriculture 2025 — Data

来源：https://www.kaggle.com/competitions/beyond-visible-spectrum-ai-for-agriculture-2025p2/data

本地 Meta Kaggle 快照；不是本次实时页面。

# Dataset Description

Task Description

Detailed submission in 
https://www.kaggle.com/code/eenysh/fid-computing-and-submission?kernelSessionId=238182542

Participants are required to develop the generative AI that generate the hyperspectral images of disease infected crop by prompt of given disease ratio.

The training data and evaluation data are classified into 10 disease level, percentage intervals of 10% each:

· Level 1: 0–10%

· Level 2: 10–20%

· Level 3: 20–30%

· ...

· Level 10: 90–100%

For each disease level, participants must:

2. Generate 50 new images corresponding to the number of evaluation images, maintaining the target class characteristics.

3. Calculate the Fréchet Inception Distance (FID) between the generated images and the evaluation images for each class.

Submission Format

Submission Format

Participants must submit a CSV file containing:

· 10 FID values corresponding to 10 disease level.

· Columns should be named as Prediction

· Values must be real numbers (float).

ID	Prediction
1	FID
2	FID
3	FID
4	FID
5	FID
6	FID
7	FID
8	FID
9	FID
10	FID
11	FID
12	FID
13	FID
14	FID
15	FID
16	FID
17	FID
18	FID
19	FID
20	FID