First commit of LightlyTrain app

app.py

@@ -0,0 +1,110 @@
+import gradio as gr
+import numpy as np
+import supervision as sv
+from PIL import Image
+import lightly_train  # Ensure this matches the installed package name
+
+# --- CONFIGURATION ---
+# We use a default LightlyTrain model so the Space works immediately.
+DEFAULT_MODEL_NAME = "dinov3/vitt16-ltdetr-coco"
+
+# --- HELPER FUNCTIONS ---
+def load_lightly_model(model_name):
+    print(f"Loading model: {model_name}...")
+    # This automatically downloads the pretrained model from Lightly
+    return lightly_train.load_model(model_name)
+
+# Initialize model once at startup to save time
+model = load_lightly_model(DEFAULT_MODEL_NAME)
+
+def predict_and_annotate(image, confidence_threshold, model_name):
+    # 1. Run Prediction using LightlyTrain
+    # LightlyTrain's predict method handles PIL images directly
+    results = model.predict(image)
+    
+    # LightlyTrain returns a dictionary: {'bboxes': Tensor, 'labels': Tensor, 'scores': Tensor}
+    # We move tensors to CPU and convert to numpy for Supervision
+    boxes = results['bboxes'].cpu().numpy()
+    labels = results['labels'].cpu().numpy()
+    scores = results['scores'].cpu().numpy()
+
+    # 2. Filter by Confidence
+    valid_indices = scores > confidence_threshold
+    boxes = boxes[valid_indices]
+    labels = labels[valid_indices]
+    scores = scores[valid_indices]
+
+    # 3. Convert to Supervision Detections format
+    detections = sv.Detections(
+        xyxy=boxes,
+        confidence=scores,
+        class_id=labels
+    )
+
+    # 4. Annotate the Image
+    box_annotator = sv.BoxAnnotator()
+    label_annotator = sv.LabelAnnotator()
+
+    # Create label text (e.g., "Class: 0 0.85")
+    # Note: If you have a class names list (like COCO_CLASSES), you can map IDs to names here.
+    generated_labels = [
+        f"Class {class_id} {confidence:.2f}"
+        for class_id, confidence in zip(detections.class_id, detections.confidence)
+    ]
+
+    annotated_image = image.copy()
+    annotated_image = box_annotator.annotate(scene=annotated_image, detections=detections)
+    annotated_image = label_annotator.annotate(scene=annotated_image, detections=detections, labels=generated_labels)
+
+    return annotated_image
+
+# --- GRADIO UI ---
+with gr.Blocks() as demo:
+    gr.Markdown("# LightlyTrain Object Detection Demo 🚀")
+    gr.Markdown("This demo uses **LightlyTrain** with a **DINOv3** backbone to detect objects.")
+    
+    with gr.Row():
+        with gr.Column():
+            input_img = gr.Image(type="pil", label="Input Image")
+            
+            conf_slider = gr.Slider(
+                minimum=0.0, 
+                maximum=1.0, 
+                value=0.3, 
+                step=0.05, 
+                label="Confidence Threshold"
+            )
+            
+            # Dropdown for model selection (currently just one default)
+            model_selector = gr.Dropdown(
+                choices=[DEFAULT_MODEL_NAME], 
+                value=DEFAULT_MODEL_NAME, 
+                label="Model Checkpoint"
+            )
+            
+            run_btn = gr.Button("Run Detection", variant="primary")
+        
+        with gr.Column():
+            output_img = gr.Image(label="Annotated Result")
+
+    # Connect the button to the function
+    run_btn.click(
+        fn=predict_and_annotate, 
+        inputs=[input_img, conf_slider, model_selector], 
+        outputs=output_img
+    )
+
+    # Example images for quick testing
+    gr.Examples(
+        examples=[
+            ["https://media.roboflow.com/notebooks/examples/dog-2.jpeg", 0.3, DEFAULT_MODEL_NAME],
+            ["https://media.roboflow.com/supervision/image-examples/vehicles.png", 0.3, DEFAULT_MODEL_NAME]
+        ],
+        inputs=[input_img, conf_slider, model_selector],
+        outputs=output_img,
+        fn=predict_and_annotate,
+        cache_examples=True,
+    )
+
+if __name__ == "__main__":
+    demo.launch()

requirements.txt

@@ -0,0 +1,7 @@
+lightly-train
+supervision
+gradio
+torch
+torchvision
+numpy
+Pillow