# Step 6: Train + Evaluate Train a 2D detector on your labeled data and evaluate it properly. Understanding **where** the model fails tells you what to label next. > **Note:** We train on `human_detections` (2D labels from Step 4) using the **left camera slice**. For evaluation, we need labels on the val set too. ## Get Training Data **Important:** We train on the **left camera slice** of annotated groups. Only samples with actual labels are included. ## Export for Training Export the left camera images and labels in YOLO format. ## Train YOLOv8 ## Run Inference on Validation ## Evaluate ## Analyze Failures Understanding failures is more important than the mAP number. In the App: 1. Filter by `failure:high_fn` to see where model missed objects 2. Filter by `failure:high_fp` to see where model hallucinated 3. Look for patterns: specific classes? distances? occlusions? These patterns tell you what to label next. ## Summary You trained and evaluated a 2D detector: - Used **left camera slice** from annotated groups - Exported in YOLO format - Trained YOLOv8n for 10 epochs - Evaluated with FiftyOne: mAP + per-sample FP/FN - Tagged failure cases for next iteration **Key insight:** The failure tags tell you what to label next. **Artifacts:** - `predictions` field with model outputs - `eval_v0` evaluation results - Failure tags: `failure:high_fn`, `failure:high_fp` - `eval_v0_failures` saved view **Next:** Step 7 - Iteration Loop