Fix TF Lite exports

[glenn-jocher]

Signed-off-by: Glenn Jocher glenn.jocher@ultralytics.com

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Improvements to documentation or project management with no code changes.

📊 Key Changes

• No actual code changes were introduced in this PR.
• Possible updates to documentation, comments, or non-code files in the repo.

🎯 Purpose & Impact

• Enhance understanding for developers and users by improving non-code content.
• Maintain project organization, possibly preparing for future code changes.
• No direct impact on the software's functionality, performance, or user experience.

[glenn-jocher]

Resolves

!python detect.py --weights yolov5s-fp16.tflite

detect: weights=['yolov5s-fp16.tflite'], source=data/images, data=data/coco128.yaml, imgsz=[640, 640], conf_thres=0.25, iou_thres=0.45, max_det=1000, device=, view_img=False, save_txt=False, save_conf=False, save_crop=False, nosave=False, classes=None, agnostic_nms=False, augment=False, visualize=False, update=False, project=runs/detect, name=exp, exist_ok=False, line_thickness=3, hide_labels=False, hide_conf=False, half=False, dnn=False, vid_stride=1
YOLOv5 🚀 v6.2-146-ge233c03 Python-3.7.14 torch-1.12.1+cu113 CUDA:0 (Tesla V100-SXM2-16GB, 16160MiB)

Loading yolov5s-fp16.tflite for TensorFlow Lite inference...
INFO: Created TensorFlow Lite XNNPACK delegate for CPU.
Traceback (most recent call last):
  File "detect.py", line 255, in <module>
    main(opt)
  File "detect.py", line 250, in main
    run(**vars(opt))
  File "/usr/local/lib/python3.7/dist-packages/torch/autograd/grad_mode.py", line 27, in decorate_context
    return func(*args, **kwargs)
  File "detect.py", line 126, in run
    pred = non_max_suppression(pred, conf_thres, iou_thres, classes, agnostic_nms, max_det=max_det)
  File "/content/yolov5/utils/general.py", line 870, in non_max_suppression
    conf, j = x[:, 5:mi].max(1, keepdim=True)
IndexError: max(): Expected reduction dim 1 to have non-zero size.

[zhiqwang]

@glenn-jocher Seems like nothing has been updated here in this PR.

[glenn-jocher]

@zhiqwang yes, you're right! I forgot to add the main change, added in #9518

TFLite models were outputting inference in different orders on some systems, so this removes training outputs on export, leaving a single detection output.