在github下载了YOLOv5的官方代码后使用官方的自动迁移推理导致速度很慢

设备：Atlas  800 (3000）

代码：yolov5.7   官方代码

官方的自动迁移：（推荐）自动迁移-模型脚本迁移-迁移适配-模型开发-Ascend Extension for PyTorch6.0.RC2开发文档-昇腾社区 (hiascend.com)

版本选择：

pytoch：2.1.0
torch_npu:6.0.rc1
python:3.9
架构：AArch64
CANN:8.0.rc1

问题：1.速度 变的很慢 一张图片的推理要14s

2.出现了 NMS的warning   看了之前论坛的内容 加了 import torchvision_npu（有nms代码的地方） 还是超时了

(test) [root@192 ~]# python "/home/yolov5-7.0/detect.py"
/root/anaconda3/envs/test/lib/python3.9/site-packages/torch_npu/contrib/transfer_to_npu.py:211: ImportWarning:
    *************************************************************************************************************
    The torch.Tensor.cuda and torch.nn.Module.cuda are replaced with torch.Tensor.npu and torch.nn.Module.npu now..
    The torch.cuda.DoubleTensor is replaced with torch.npu.FloatTensor cause the double type is not supported now..
    The backend in torch.distributed.init_process_group set to hccl now..
    The torch.cuda.* and torch.cuda.amp.* are replaced with torch.npu.* and torch.npu.amp.* now..
    The device parameters have been replaced with npu in the function below:
    torch.logspace, torch.randint, torch.hann_window, torch.rand, torch.full_like, torch.ones_like, torch.rand_like, torch.randperm, torch.arange, torch.frombuffer, torch.normal, torch._empty_per_channel_affine_quantized, torch.empty_strided, torch.empty_like, torch.scalar_tensor, torch.tril_indices, torch.bartlett_window, torch.ones, torch.sparse_coo_tensor, torch.randn, torch.kaiser_window, torch.tensor, torch.triu_indices, torch.as_tensor, torch.zeros, torch.randint_like, torch.full, torch.eye, torch._sparse_csr_tensor_unsafe, torch.empty, torch._sparse_coo_tensor_unsafe, torch.blackman_window, torch.zeros_like, torch.range, torch.sparse_csr_tensor, torch.randn_like, torch.from_file, torch._cudnn_init_dropout_state, torch._empty_affine_quantized, torch.linspace, torch.hamming_window, torch.empty_quantized, torch._pin_memory, torch.autocast, torch.load, torch.Generator, torch.Tensor.new_empty, torch.Tensor.new_empty_strided, torch.Tensor.new_full, torch.Tensor.new_ones, torch.Tensor.new_tensor, torch.Tensor.new_zeros, torch.Tensor.to, torch.nn.Module.to, torch.nn.Module.to_empty
    *************************************************************************************************************

warnings.warn(msg, ImportWarning)
/root/anaconda3/envs/test/lib/python3.9/site-packages/thop/profile.py:12: DeprecationWarning: distutils Version classes are deprecated. Use packaging.version instead.
  if LooseVersion(torch.__version__) < LooseVersion("1.0.0"):
detect: weights=../home/yolov5-7.0/yolov5s.pt, source=../home/yolov5-7.0/data/images, data=../home/yolov5-7.0/data/coco128.yaml, imgsz=[640, 640], conf_thres=0.25, iou_thres=0.45, max_det=1000, device=npu, 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=../home/yolov5-7.0/runs/detect, name=exp, exist_ok=False, line_thickness=3, hide_labels=False, hide_conf=False, half=False, dnn=False, vid_stride=1
YOLOv5 í ½íº€ 1.0 torch 2.1.0 NPU Ascend310P3

Fusing layers...
YOLOv5s summary: 213 layers, 7225885 parameters, 0 gradients, 16.4 GFLOPs
[W OpCommand.cpp:88] Warning: [Check][offset] Check input storage_offset[%ld] = 0 failed, result is untrustworthy5 (function operator())
WARNING ⚠️ NMS time limit 0.550s exceeded
image 1/2 /home/yolov5-7.0/data/images/bus.jpg: 640x480 4 persons, 1 bus, 14503.2ms
WARNING ⚠️ NMS time limit 0.550s exceeded
image 2/2 /home/yolov5-7.0/data/images/zidane.jpg: 384x640 2 persons, 2 ties, 14584.1ms
Speed: 252.7ms pre-process, 14543.6ms inference, 5094.1ms NMS per image at shape (1, 3, 640, 640)
Results saved to ../home/yolov5-7.0/runs/detect/exp12

Ascend/modelzoo
暂停

内容风险标识

评论 (0)

Ascend/modelzoo暂停 .gitee-modal { width: 500px !important; }

内容风险标识