Practical Guidance - Practical Guidance - 6.3 English - XD056

Set has_objectness_score based on model architecture, not the model’s internal precision (INT8/FP16/FP32). Mixed-precision compiled models still follow the same output layout as their architecture defines.

For YOLOv8 and later anchor-free models, set has_objectness_score to false and box_grid_decode to false.

For YOLOv3/v5 anchor-based models, set has_objectness_score to true. Enable box_grid_decode only if the model outputs raw grid-relative predictions.

If using a YOLO preset, remember that explicitly configured JSON fields override the preset defaults.

When box_grid_decode is true, configure input_width, input_height, grid_strides, and preds_per_location so the generated grid count matches the model’s prediction count. For example, a P6 model may use [8, 16, 32, 64], and a P2-P5 model may use [4, 8, 16, 32].

INT8 tensors are automatically dequantized using the scale coefficient from the tensor info provided by the inference engine.