TensorRT converted yolov5 has multiple outputs

[bdv29]

So I converted a custom trained model to TensorRT to run on the Jetson nano,
and the output I get is a list with four arrays,
the fourth array seems to be the one that gives me the correct bounding boxes when reshaped correctly taking account the amount of classes I have trained it for and fed into the NMS function ,
but I'm wondering is what are the other three arrays for? Is it just some odd idiosyncrasy of tensorRT or is there another aspect of yolo that I don't understand and could those other arrays have useful information?

screenshot of the four arrays

[glenn-jocher]

@bdv29 0, 1 and 2 in your screenshot are part of 3. See TensorRT Deployment Tutorial for exporting to TensorRT:

YOLOv5 Tutorials

• Train Custom Data  🚀 RECOMMENDED
• Tips for Best Training Results  ☘️ RECOMMENDED
• Weights & Biases Logging  🌟 NEW
• Supervisely Ecosystem  🌟 NEW
• Multi-GPU Training
• PyTorch Hub  ⭐ NEW
• TorchScript, ONNX, CoreML Export 🚀
• Test-Time Augmentation (TTA)
• Model Ensembling
• Model Pruning/Sparsity
• Hyperparameter Evolution
• Transfer Learning with Frozen Layers  ⭐ NEW
• TensorRT Deployment