File size: 15,292 Bytes
e5c06a8
 
 
 
 
 
 
 
 
 
 
4db6d78
e5c06a8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a7559f5
 
 
 
 
 
 
 
 
e5c06a8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4db6d78
 
e5c06a8
 
 
 
 
a7559f5
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
"""
PaddockAI β€” Livestock & Pasture Assessment Tool
Snap a photo on your phone β†’ get instant ML-powered analysis.
Uses Qwen3-VL-8B vision-language model via HF Inference API.
"""

import os
import io
import re
import base64
import datetime
import gradio as gr
from PIL import Image
from huggingface_hub import InferenceClient

# ---------------------------------------------------------------------------
# Model & provider config β€” tries multiple providers for resilience
# ---------------------------------------------------------------------------
MODEL_ID = "Qwen/Qwen3-VL-8B-Instruct"
PROVIDERS = ["novita", "together", "auto"]


def call_vlm(messages: list, max_tokens: int = 1024) -> str:
    """Call VLM with automatic provider failover."""
    token = os.environ.get("HF_TOKEN")
    last_error = None

    for provider in PROVIDERS:
        try:
            client = InferenceClient(provider=provider, api_key=token)
            response = client.chat.completions.create(
                model=MODEL_ID,
                messages=messages,
                max_tokens=max_tokens,
                temperature=0.3,
            )
            result = response.choices[0].message.content
            # Strip chain-of-thought <think> blocks from Qwen3
            if "<think>" in result:
                result = re.sub(r"<think>.*?</think>", "", result, flags=re.DOTALL).strip()
            return result
        except Exception as e:
            last_error = e
            continue

    raise last_error or RuntimeError("All providers failed")


# ---------------------------------------------------------------------------
# Expert agricultural prompts
# ---------------------------------------------------------------------------
SYSTEM_PROMPT = (
    "You are PaddockAI, an expert livestock and pasture assessment assistant "
    "for farmers, feedlotters, and graziers worldwide.\n\n"
    "Rules:\n"
    "- Be specific: give numbers, scores, and concrete recommendations.\n"
    "- Use industry-standard scoring systems (BCS 1-5, pasture condition 1-5).\n"
    "- If you cannot determine something from the image, say so honestly.\n"
    "- Format with markdown headers and bullet points for phone readability.\n"
    "- Keep responses practical β€” these people are in the paddock, not at a desk."
)

TASK_PROMPTS = {
    "πŸ„ Cattle Health Check": """Perform a thorough visual health assessment of the cattle in this image.

**Assess and report on:**
1. **Body Condition Score (BCS)** β€” 1–5 scale (1=emaciated, 3=good, 5=obese). Note fat cover over ribs, spine, hips, tailhead.
2. **Coat & Skin** β€” Condition, shine, lesions, lumps, hair loss, external parasites (lice, ticks, mange).
3. **Eyes & Nose** β€” Discharge, swelling, cloudiness, pinkeye/IBR signs.
4. **Gait & Posture** β€” Lameness, stiffness, favouring a leg, abnormal stance.
5. **Respiratory** β€” Laboured breathing, nasal discharge, coughing posture.
6. **General Demeanour** β€” Alert vs dull, separated from mob, ear position.

**Provide:**
- Overall health rating: βœ… Healthy / ⚠️ Monitor / 🚨 Needs Attention
- Specific concerns identified
- Recommended actions (draft for inspection, call vet, treat for X)

If multiple animals are visible, assess the mob's general condition and flag any individuals of concern.""",

    "🌿 Pasture Condition": """Assess the pasture/paddock condition shown in this image.

**Evaluate:**
1. **Ground Cover (%)** β€” Vegetation vs bare soil.
2. **Feed on Offer (FOO)** β€” Estimate kg DM/ha. <500 very low, 500-1000 low, 1000-2000 moderate, 2000-3000 good, 3000+ high.
3. **Pasture Composition** β€” Grass species (phalaris, ryegrass, natives, clover, lucerne). Desirable vs undesirable ratio.
4. **Weed Burden** β€” Identify weeds, estimate severity (low/moderate/high).
5. **Pasture Quality** β€” Green vs dry/hayed-off, growth stage (vegetative, reproductive, senescent).
6. **Grazing Pressure** β€” Over-grazing signs (short, bare patches, pugging) or under-grazing (rank, gone to seed).
7. **Soil Condition** β€” Erosion, waterlogging, compaction, pugging damage.

**Provide:**
- Pasture condition score: 1 (very poor) to 5 (excellent)
- Estimated carrying capacity (DSE/ha or head/ha)
- Top 3 management recommendations""",

    "πŸ“Š Body Condition Score": """Perform a detailed Body Condition Score (BCS) assessment.

**1–5 scale (half scores OK):**
| Score | Description |
|-------|-------------|
| 1.0 | Emaciated β€” bones prominent, no fat cover |
| 2.0 | Thin β€” some bone visible, slight cover |
| 3.0 | Good store β€” ribs felt with firm pressure |
| 4.0 | Fat β€” ribs cannot be felt, fat deposits visible |
| 5.0 | Obese β€” heavy fat deposits, mobility affected |

**For each animal, describe:**
- Fat cover: spine, ribs, hips/hooks, tailhead, brisket
- Muscle development: loin, rump, shoulder
- Breed type if identifiable
- BCS score with justification

**Then provide:**
- Target BCS for the likely class (breeder, feeder, store)
- Whether current BCS is on track or needs action
- Feed/management recommendations to reach target""",

    "πŸ”’ Count Stock": """Count the livestock visible in this image as accurately as possible.

**Report:**
1. **Total count** β€” Exact number or confident range (e.g., 45-50 head).
2. **Method** β€” How you counted (sections, rows, clusters).
3. **Confidence** β€” High/Medium/Low.
4. **Species & class** β€” Cattle (cows, calves, steers, bulls), sheep, goats, etc.
5. **Mob density** β€” Too tight (welfare concern) or well-spaced.
6. **Separated animals** β€” Any away from mob (sick, calving, or needing check).

For drone/aerial shots, estimate area covered and density.""",

    "πŸ₯ Injury / Lameness": """Assess this image for signs of injury, lameness, or physical ailment.

**Look for:**
1. **Lameness** β€” Favouring a leg, head bob, shortened stride, reluctance to bear weight, swollen joints.
2. **Injuries** β€” Cuts, wounds, swelling, abscesses, broken horns, eye injuries, prolapse.
3. **Skin conditions** β€” Ringworm, warts, lumps, cancer eye, sunburn, photosensitisation.
4. **Structural** β€” Poor feet, overgrown hooves, down animal, difficulty rising.
5. **Dystocia signs** β€” If calving/lambing: progress, presentation, time concerns.

**Provide:**
- Urgency: 🟒 Minor / 🟑 Moderate β€” monitor closely / πŸ”΄ Urgent β€” immediate attention
- Likely cause or differentials
- Recommended immediate action
- Whether a vet call is warranted

⚠️ **Visual assessment only β€” not a veterinary diagnosis. Always consult your vet for serious concerns.**""",

    "🌾 General / Other": """Provide a comprehensive agricultural assessment of this image.

Describe what you see: livestock, pasture, infrastructure, landscape, weather conditions.
Identify anything noteworthy: potential issues, good management practices, areas for improvement.
Give practical recommendations.""",
}


# ---------------------------------------------------------------------------
# Image processing
# ---------------------------------------------------------------------------
def image_to_base64(image: Image.Image, max_size: int = 1280) -> str:
    """Convert PIL Image to base64 JPEG, resizing if needed."""
    if image.mode == "RGBA":
        image = image.convert("RGB")
    w, h = image.size
    if max(w, h) > max_size:
        ratio = max_size / max(w, h)
        image = image.resize((int(w * ratio), int(h * ratio)), Image.LANCZOS)
    buf = io.BytesIO()
    image.save(buf, format="JPEG", quality=85)
    return base64.b64encode(buf.getvalue()).decode("utf-8")


# ---------------------------------------------------------------------------
# Core analysis function
# ---------------------------------------------------------------------------
def analyze(image, task: str, custom_notes: str, history: list):
    """Run VLM analysis on the uploaded image."""
    if image is None:
        return "πŸ“· **Please upload or take a photo first.**", history

    prompt = TASK_PROMPTS[task]
    if custom_notes and custom_notes.strip():
        prompt += f"\n\n**Additional context from the user:** {custom_notes.strip()}"

    b64 = image_to_base64(image)

    try:
        messages = [
            {"role": "system", "content": SYSTEM_PROMPT},
            {
                "role": "user",
                "content": [
                    {"type": "image_url", "image_url": {"url": f"data:image/jpeg;base64,{b64}"}},
                    {"type": "text", "text": prompt},
                ],
            },
        ]
        result = call_vlm(messages)

        # Add to history
        timestamp = datetime.datetime.now().strftime("%H:%M:%S")
        entry = f"### {task} β€” {timestamp}\n\n{result}\n\n---\n"
        history = history or []
        history.insert(0, entry)

        return result, history

    except Exception as e:
        error_msg = str(e)
        if "rate" in error_msg.lower() or "limit" in error_msg.lower():
            msg = "⏳ **Rate limit reached.** Please wait a moment and try again."
        elif "401" in error_msg or "auth" in error_msg.lower():
            msg = "πŸ”‘ **Authentication error.** The service is temporarily unavailable."
        else:
            msg = f"❌ **Analysis failed:** {error_msg}\n\nPlease try again."
        return msg, history


def format_history(history: list) -> str:
    """Format assessment history for display."""
    if not history:
        return "*No assessments yet. Upload a photo to get started.*"
    return "\n".join(history[:10])  # Keep last 10


# ---------------------------------------------------------------------------
# UI
# ---------------------------------------------------------------------------
HEADER_MD = """
# πŸ„ PaddockAI
### Instant livestock & pasture assessment β€” right from your phone

**How to use:** Upload or snap a photo β†’ Pick an assessment type β†’ Tap **Analyze**

> ⚠️ Not a substitute for veterinary advice. For health concerns, always consult your vet.
"""

CSS = """
/* Big touch targets for use in the paddock */
.primary-btn { min-height: 56px !important; font-size: 18px !important; font-weight: 600 !important; }
.task-radio label { padding: 10px 14px !important; margin: 3px 0 !important; border-radius: 8px !important; }
.output-box { font-size: 15px !important; line-height: 1.65 !important; }

@media (max-width: 768px) {
    .primary-btn { min-height: 64px !important; font-size: 20px !important; }
    .contain { padding: 6px !important; }
}
"""


THEME = gr.themes.Soft(
    primary_hue="green",
    secondary_hue="emerald",
    neutral_hue="stone",
)


def build_ui():
    with gr.Blocks() as demo:

        # State for assessment history
        history_state = gr.State([])

        gr.Markdown(HEADER_MD)

        with gr.Tabs():
            # ── Tab 1: Assess ────────────────────────────────────────
            with gr.Tab("πŸ“· Assess", id="assess"):
                with gr.Row(equal_height=False):
                    with gr.Column(scale=1, min_width=300):
                        image_input = gr.Image(
                            type="pil",
                            sources=["upload", "webcam", "clipboard"],
                            label="πŸ“· Photo",
                            height=350,
                        )
                        task_selector = gr.Radio(
                            choices=list(TASK_PROMPTS.keys()),
                            value="πŸ„ Cattle Health Check",
                            label="Assessment Type",
                            elem_classes=["task-radio"],
                        )
                        custom_notes = gr.Textbox(
                            label="Notes (optional)",
                            placeholder="e.g., 'Heifer off feed 2 days' or 'Paddock rested 3 weeks'",
                            lines=2,
                        )
                        analyze_btn = gr.Button(
                            "πŸ” Analyze",
                            variant="primary",
                            size="lg",
                            elem_classes=["primary-btn"],
                        )

                    with gr.Column(scale=1, min_width=300):
                        output = gr.Markdown(
                            value="*Upload a photo and tap Analyze to get started.*",
                            elem_classes=["output-box"],
                        )

            # ── Tab 2: History ───────────────────────────────────────
            with gr.Tab("πŸ“‹ History", id="history"):
                history_display = gr.Markdown(
                    value="*No assessments yet.*",
                    elem_classes=["output-box"],
                )

            # ── Tab 3: Tips ──────────────────────────────────────────
            with gr.Tab("πŸ’‘ Tips", id="tips"):
                gr.Markdown("""
## πŸ“Έ Getting the best results

### Cattle photos
- **Side-on** is best for body condition scoring
- Get close enough to see **eyes and nose** for health checks
- **Multiple animals?** Try to get them all clearly visible
- For mob assessments, step back for a wider shot

### Pasture / paddock
- Include some **sky or horizon** for scale reference
- Shoot at **slight downward angle** to show ground cover
- Multiple angles from different spots give better estimates

### Counting stock
- **Elevated positions** (hill, vehicle roof, drone) work best
- Try to capture the **whole mob** in one frame
- Mark the image mentally in sections for counting

### Injury / lameness
- Get as **close and clear** as safely possible
- Photograph from **multiple angles**
- Include a **size reference** if possible (hand, boot, etc.)

### General tips
- β˜€οΈ **Natural daylight** gives the best results
- πŸ“± Clean your phone lens!
- πŸ”„ If the first result seems off, try a different angle or assessment type
                """)

        # Footer
        gr.Markdown(
            "---\n"
            "*Powered by [Qwen3-VL-8B](https://huggingface.co/Qwen/Qwen3-VL-8B-Instruct) "
            "via [HF Inference API](https://huggingface.co/docs/inference-providers) Β· "
            "Built for farmers, by farmers*"
        )

        # ── Wiring ───────────────────────────────────────────────
        analyze_btn.click(
            fn=analyze,
            inputs=[image_input, task_selector, custom_notes, history_state],
            outputs=[output, history_state],
            show_progress="minimal",
        ).then(
            fn=format_history,
            inputs=[history_state],
            outputs=[history_display],
        )

    return demo


# ---------------------------------------------------------------------------
# Launch
# ---------------------------------------------------------------------------
if __name__ == "__main__":
    demo = build_ui()
    demo.launch(
        server_name="0.0.0.0",
        server_port=7860,
        theme=THEME,
        css=CSS,
    )