Initial upload: Gradio app for Pseudo-nitzschia cell counting

README.md

@@ -1,12 +1,21 @@
 ---
-title: Pseudo Nitzschia Counter
-emoji: 😻
-colorFrom: indigo
-colorTo: blue
+title: Pseudo-nitzschia Cell Counter
+emoji: 🔬
+colorFrom: blue
+colorTo: green
 sdk: gradio
-sdk_version: 6.9.0
+sdk_version: "5.23.0"
 app_file: app.py
 pinned: false
+license: mit
+models:
+  - patcdaniel/pseudo-nitzschia-yolo11s-seg
+datasets:
+  - patcdaniel/pseudo-nitzschia-cell-segmentation
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# Pseudo-nitzschia Cell Counter
+
+Drag and drop an IFCB image of a *Pseudo-nitzschia* chain colony to automatically count individual cells using instance segmentation.
+
+**Model:** [YOLO11s-seg](https://huggingface.co/patcdaniel/pseudo-nitzschia-yolo11s-seg) fine-tuned on [302 annotated IFCB images](https://huggingface.co/datasets/patcdaniel/pseudo-nitzschia-cell-segmentation).

app.py

@@ -0,0 +1,147 @@
+"""
+Pseudo-nitzschia Cell Counter — Gradio Space
+Drag-and-drop an IFCB image to count individual cells using YOLO11s instance segmentation.
+"""
+
+import gradio as gr
+import numpy as np
+import cv2
+from PIL import Image
+from ultralytics import YOLO
+from huggingface_hub import hf_hub_download
+import matplotlib.pyplot as plt
+
+# Download and load model at startup
+MODEL_REPO = "patcdaniel/pseudo-nitzschia-yolo11s-seg"
+model_path = hf_hub_download(MODEL_REPO, "best.pt")
+model = YOLO(model_path)
+
+
+def count_cells(input_image, confidence_threshold, iou_threshold):
+    """Run inference on an uploaded image and return annotated result with cell count."""
+    if input_image is None:
+        return None, "No image provided."
+
+    # Convert PIL to numpy BGR for YOLO
+    img_rgb = np.array(input_image)
+    if len(img_rgb.shape) == 2:
+        # Grayscale -> RGB
+        img_rgb = cv2.cvtColor(img_rgb, cv2.COLOR_GRAY2RGB)
+    elif img_rgb.shape[2] == 4:
+        # RGBA -> RGB
+        img_rgb = cv2.cvtColor(img_rgb, cv2.COLOR_RGBA2RGB)
+
+    # Run inference
+    results = model.predict(
+        source=img_rgb,
+        conf=confidence_threshold,
+        iou=iou_threshold,
+        imgsz=640,
+        verbose=False,
+    )
+
+    result = results[0]
+
+    # Check for detections
+    if result.masks is None or len(result.masks) == 0:
+        return input_image, "**0 cells detected.**"
+
+    masks = result.masks.data.cpu().numpy()
+    confidences = result.boxes.conf.cpu().numpy()
+    cell_count = len(masks)
+
+    # Create visualization
+    overlay = img_rgb.copy()
+    colors = plt.cm.rainbow(np.linspace(0, 1, cell_count))
+
+    polygons = []
+    areas = []
+    for i, mask in enumerate(masks):
+        mask_uint8 = (mask * 255).astype(np.uint8)
+        # Resize mask to image dimensions if needed
+        if mask_uint8.shape[:2] != img_rgb.shape[:2]:
+            mask_uint8 = cv2.resize(mask_uint8, (img_rgb.shape[1], img_rgb.shape[0]))
+
+        contours, _ = cv2.findContours(mask_uint8, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
+        if contours:
+            largest = max(contours, key=cv2.contourArea)
+            epsilon = 0.001 * cv2.arcLength(largest, True)
+            poly = cv2.approxPolyDP(largest, epsilon, True)
+            color = (colors[i][:3] * 255).astype(np.uint8).tolist()
+
+            cv2.fillPoly(overlay, [poly], color)
+            cv2.polylines(img_rgb, [poly], True, color, 2)
+
+            area = cv2.contourArea(largest)
+            areas.append(area)
+            polygons.append(poly)
+
+    # Blend overlay
+    alpha = 0.3
+    annotated = cv2.addWeighted(overlay, alpha, img_rgb, 1 - alpha, 0)
+
+    # Add cell count label
+    text = f"Cells: {cell_count}"
+    font = cv2.FONT_HERSHEY_SIMPLEX
+    font_scale = max(0.6, min(img_rgb.shape[1] / 600, 1.5))
+    thickness = max(1, int(font_scale * 2))
+    (tw, th), _ = cv2.getTextSize(text, font, font_scale, thickness)
+    cv2.rectangle(annotated, (5, 5), (15 + tw, 15 + th), (255, 255, 255), -1)
+    cv2.putText(annotated, text, (10, 10 + th), font, font_scale, (0, 0, 200), thickness)
+
+    # Build summary text
+    avg_conf = float(np.mean(confidences))
+    avg_area = float(np.mean(areas)) if areas else 0
+
+    summary = (
+        f"### {cell_count} cell{'s' if cell_count != 1 else ''} detected\n\n"
+        f"| Metric | Value |\n"
+        f"|--------|-------|\n"
+        f"| Cell count | {cell_count} |\n"
+        f"| Avg. confidence | {avg_conf:.3f} |\n"
+        f"| Confidence range | {float(confidences.min()):.3f} – {float(confidences.max()):.3f} |\n"
+        f"| Avg. mask area (px) | {avg_area:.1f} |\n"
+    )
+
+    return Image.fromarray(annotated), summary
+
+
+# Build Gradio interface
+with gr.Blocks(title="Pseudo-nitzschia Cell Counter") as demo:
+    gr.Markdown(
+        "# Pseudo-nitzschia Cell Counter\n"
+        "Upload an IFCB image of a *Pseudo-nitzschia* chain colony to count individual cells "
+        "using [YOLO11s instance segmentation](https://huggingface.co/patcdaniel/pseudo-nitzschia-yolo11s-seg)."
+    )
+
+    with gr.Row():
+        with gr.Column():
+            input_image = gr.Image(type="pil", label="Upload Image")
+            with gr.Row():
+                conf_slider = gr.Slider(
+                    minimum=0.05, maximum=0.95, value=0.25, step=0.05,
+                    label="Confidence Threshold"
+                )
+                iou_slider = gr.Slider(
+                    minimum=0.1, maximum=0.95, value=0.7, step=0.05,
+                    label="IoU Threshold (NMS)"
+                )
+            run_btn = gr.Button("Count Cells", variant="primary")
+
+        with gr.Column():
+            output_image = gr.Image(type="pil", label="Segmentation Result")
+            output_text = gr.Markdown(label="Results")
+
+    run_btn.click(
+        fn=count_cells,
+        inputs=[input_image, conf_slider, iou_slider],
+        outputs=[output_image, output_text],
+    )
+
+    gr.Markdown(
+        "---\n"
+        "**Model:** [patcdaniel/pseudo-nitzschia-yolo11s-seg](https://huggingface.co/patcdaniel/pseudo-nitzschia-yolo11s-seg) | "
+        "**Dataset:** [patcdaniel/pseudo-nitzschia-cell-segmentation](https://huggingface.co/datasets/patcdaniel/pseudo-nitzschia-cell-segmentation)"
+    )
+
+demo.launch()

requirements.txt

@@ -0,0 +1,5 @@
+ultralytics>=8.0.0
+huggingface_hub>=0.20.0
+numpy>=1.24.0
+opencv-python-headless>=4.8.0
+pillow>=10.0.0