Crop Disease Detection Images

1
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
2
# SPDX-License-Identifier: Apache-2.0
3
# /// script
4
# requires-python = ">=3.10"
5
# dependencies = [
6
#     "data-designer",
7
# ]
8
# ///
9
"""Agriculture Crop Disease Detection Image Recipe
10
11
Generate synthetic crop disease detection images with controlled variation over
12
crop type, growth stage, viewpoint, disease or confounding condition, severity,
13
weather, irrigation, and field condition. The objective is to create examples
14
where the expected crop-health label is known, including healthy negatives and
15
hard confounders, so teams can evaluate detection prompts, build labeling
16
rubrics, calibrate reviewers, and prototype crop-disease workflows before using
17
governed field imagery.
18
19
Prerequisites:
20
    - An image-generation provider key for the selected model. The defaults use
21
      OpenRouter, so set OPENROUTER_API_KEY before running.
22
23
Run:
24
    uv run agriculture_crop_imagery.py --num-records 10
25
"""
26
27
from __future__ import annotations
28
29
import argparse
30
from pathlib import Path
31
32
import data_designer.config as dd
33
from data_designer.interface import DataDesigner, DatasetCreationResults
34
35
DEFAULT_MODEL_PROVIDER = "openrouter"
36
DEFAULT_MODEL_ID = "google/gemini-3.1-flash-image-preview"
37
DEFAULT_MODEL_ALIAS = "agriculture-image-model"
38
39
40
def build_model_configs(
41
    *,
42
    model_provider: str,
43
    model_id: str,
44
    model_alias: str,
45
    image_size: str,
46
    aspect_ratio: str,
47
    max_parallel_requests: int,
48
) -> list[dd.ModelConfig]:
49
    return [
50
        dd.ModelConfig(
51
            alias=model_alias,
52
            model=model_id,
53
            provider=model_provider,
54
            inference_parameters=dd.ImageInferenceParams(
55
                extra_body={
56
                    "modalities": ["image", "text"],
57
                    "image_config": {
58
                        "aspect_ratio": aspect_ratio,
59
                        "image_size": image_size,
60
                    },
61
                },
62
                max_parallel_requests=max_parallel_requests,
63
            ),
64
            skip_health_check=True,
65
        )
66
    ]
67
68
69
def add_category(config_builder: dd.DataDesignerConfigBuilder, name: str, values: list[str]) -> None:
70
    config_builder.add_column(
71
        dd.SamplerColumnConfig(
72
            name=name,
73
            sampler_type=dd.SamplerType.CATEGORY,
74
            params=dd.CategorySamplerParams(values=values),
75
        )
76
    )
77
78
79
def add_visual_variation_id(config_builder: dd.DataDesignerConfigBuilder) -> None:
80
    """Add a unique row-level key that discourages duplicate image generations."""
81
    config_builder.add_column(
82
        dd.SamplerColumnConfig(
83
            name="visual_variation_id",
84
            sampler_type=dd.SamplerType.UUID,
85
            params=dd.UUIDSamplerParams(prefix="crop-", short_form=True),
86
        )
87
    )
88
89
90
def build_config(
91
    *,
92
    model_provider: str = DEFAULT_MODEL_PROVIDER,
93
    model_id: str = DEFAULT_MODEL_ID,
94
    model_alias: str = DEFAULT_MODEL_ALIAS,
95
    image_size: str = "1K",
96
    aspect_ratio: str = "4:3",
97
    max_parallel_requests: int = 10,
98
) -> dd.DataDesignerConfigBuilder:
99
    model_configs = build_model_configs(
100
        model_provider=model_provider,
101
        model_id=model_id,
102
        model_alias=model_alias,
103
        image_size=image_size,
104
        aspect_ratio=aspect_ratio,
105
        max_parallel_requests=max_parallel_requests,
106
    )
107
    config_builder = dd.DataDesignerConfigBuilder(model_configs=model_configs)
108
    add_visual_variation_id(config_builder)
109
110
    add_category(
111
        config_builder,
112
        "crop_type",
113
        [
114
            "corn",
115
            "soybean",
116
            "wheat",
117
            "rice",
118
            "tomato",
119
            "grape vineyard",
120
            "apple orchard",
121
            "lettuce",
122
            "potato",
123
            "strawberry",
124
        ],
125
    )
126
    add_category(
127
        config_builder,
128
        "growth_stage",
129
        [
130
            "seedling",
131
            "vegetative growth",
132
            "flowering",
133
            "fruiting",
134
            "grain fill",
135
            "near harvest",
136
        ],
137
    )
138
    add_category(
139
        config_builder,
140
        "viewpoint",
141
        [
142
            "close-up leaf-level scouting photo",
143
            "row-level field photo",
144
            "drone oblique field view",
145
            "top-down drone crop-row view",
146
            "greenhouse bench view",
147
            "orchard row view",
148
        ],
149
    )
150
    add_category(
151
        config_builder,
152
        "disease_or_condition",
153
        [
154
            "healthy crop with no visible disease",
155
            "powdery mildew on leaves",
156
            "rust-colored fungal pustules on leaf surfaces",
157
            "early blight with concentric brown leaf spots",
158
            "late blight with irregular dark lesions",
159
            "bacterial leaf spot with small dark speckles",
160
            "downy mildew patches on leaf undersides",
161
            "leaf curl with mosaic discoloration",
162
            "insect feeding damage as a disease confounder",
163
            "nutrient deficiency yellowing as a disease confounder",
164
        ],
165
    )
166
    disease_severity_values = [
167
        "low severity affecting isolated plants",
168
        "moderate severity affecting patches",
169
        "high severity affecting large field sections",
170
    ]
171
    config_builder.add_column(
172
        dd.SamplerColumnConfig(
173
            name="severity",
174
            sampler_type=dd.SamplerType.SUBCATEGORY,
175
            params=dd.SubcategorySamplerParams(
176
                category="disease_or_condition",
177
                values={
178
                    "healthy crop with no visible disease": ["none - healthy negative"],
179
                    "powdery mildew on leaves": disease_severity_values,
180
                    "rust-colored fungal pustules on leaf surfaces": disease_severity_values,
181
                    "early blight with concentric brown leaf spots": disease_severity_values,
182
                    "late blight with irregular dark lesions": disease_severity_values,
183
                    "bacterial leaf spot with small dark speckles": disease_severity_values,
184
                    "downy mildew patches on leaf undersides": disease_severity_values,
185
                    "leaf curl with mosaic discoloration": disease_severity_values,
186
                    "insect feeding damage as a disease confounder": ["confounder - not a disease severity label"],
187
                    "nutrient deficiency yellowing as a disease confounder": [
188
                        "confounder - not a disease severity label"
189
                    ],
190
                },
191
            ),
192
        )
193
    )
194
    add_category(
195
        config_builder,
196
        "field_condition",
197
        [
198
            "uniform crop stand",
199
            "patchy emergence",
200
            "uneven row spacing",
201
            "visible irrigation lines",
202
            "muddy soil after rain",
203
            "dry cracked soil",
204
            "mulched bed system",
205
        ],
206
    )
207
    add_category(
208
        config_builder,
209
        "weather_lighting",
210
        [
211
            "bright midday sun",
212
            "soft overcast light",
213
            "golden hour light",
214
            "after-rain humid conditions",
215
            "hazy smoky sky",
216
            "greenhouse diffuse lighting",
217
        ],
218
    )
219
220
    config_builder.add_column(
221
        dd.ImageColumnConfig(
222
            name="crop_image",
223
            prompt=AGRICULTURE_IMAGE_PROMPT,
224
            model_alias=model_alias,
225
        )
226
    )
227
228
    return config_builder
229
230
231
def create_dataset(
232
    config_builder: dd.DataDesignerConfigBuilder,
233
    *,
234
    num_records: int,
235
    dataset_name: str,
236
    artifact_path: Path | str | None = None,
237
) -> DatasetCreationResults:
238
    data_designer = DataDesigner(artifact_path=artifact_path)
239
    data_designer.validate(config_builder)
240
    return data_designer.create(config_builder, num_records=num_records, dataset_name=dataset_name)
241
242
243
AGRICULTURE_IMAGE_PROMPT = """\
244
Create a realistic crop disease detection image.
245
246
Scene requirements:
247
- Visual variation ID, for internal diversity only: {{ visual_variation_id }}
248
- Crop type: {{ crop_type }}
249
- Growth stage: {{ growth_stage }}
250
- Viewpoint: {{ viewpoint }}
251
- Disease or condition: {{ disease_or_condition }}
252
- Severity: {{ severity }}
253
- Field condition: {{ field_condition }}
254
- Weather and lighting: {{ weather_lighting }}
255
256
Make the image useful for crop disease detection, visual QA, reviewer
257
calibration, and data-labeling experiments. The requested crop, condition,
258
severity, and field context should be visually inspectable. Show realistic
259
plant structure, leaves, rows, soil, and disease symptoms when requested. For
260
healthy examples, show clear healthy leaves or canopy with no visible disease.
261
For confounders, make the non-disease condition plausible enough to test a
262
classifier or VLM prompt. Do not include real farm names, readable license
263
plates, watermarks, or people as the primary subject. Generate exactly one
264
final crop image for this row. Do not return alternate versions, a grid, a pair
265
of examples, before/after panels, or multiple frames. Use the visual variation
266
ID only as an internal diversity key; never render it as text.
267
"""
268
269
270
def parse_args() -> argparse.Namespace:
271
    parser = argparse.ArgumentParser(description="Generate synthetic crop disease detection imagery.")
272
    parser.add_argument("--num-records", type=int, default=10, help="Number of crop images to generate.")
273
    parser.add_argument("--dataset-name", default="crop-disease-detection-images", help="Output dataset name.")
274
    parser.add_argument("--artifact-path", type=Path, default=None, help="Optional Data Designer artifact directory.")
275
    parser.add_argument("--model-provider", default=DEFAULT_MODEL_PROVIDER, help="Image model provider name.")
276
    parser.add_argument("--model-id", default=DEFAULT_MODEL_ID, help="Provider model ID.")
277
    parser.add_argument("--model-alias", default=DEFAULT_MODEL_ALIAS, help="Alias used by image columns.")
278
    parser.add_argument("--image-size", default="1K", help="OpenRouter image size tier, such as 1K, 2K, or 4K.")
279
    parser.add_argument("--aspect-ratio", default="4:3", help="Provider-specific aspect ratio value.")
280
    parser.add_argument("--max-parallel-requests", type=int, default=10, help="Maximum parallel image requests.")
281
    return parser.parse_args()
282
283
284
def main() -> None:
285
    args = parse_args()
286
    config_builder = build_config(
287
        model_provider=args.model_provider,
288
        model_id=args.model_id,
289
        model_alias=args.model_alias,
290
        image_size=args.image_size,
291
        aspect_ratio=args.aspect_ratio,
292
        max_parallel_requests=args.max_parallel_requests,
293
    )
294
    results = create_dataset(
295
        config_builder,
296
        num_records=args.num_records,
297
        dataset_name=args.dataset_name,
298
        artifact_path=args.artifact_path,
299
    )
300
    dataset = results.load_dataset()
301
    print(f"Generated {len(dataset)} crop disease detection image rows.")
302
    print(f"Dataset artifacts: {results.artifact_storage.base_dataset_path}")
303
304
305
if __name__ == "__main__":
306
    main()