【模型评估】FastestDet

[kurisaW]

fastdet

[kurisaW]

训练日志

(yolo-fastesdet) kurisaw@hxx-System-Product-Name:~/Desktop/FastestDet-main$ python train.py 
Load yaml sucess...
<utils.tool.LoadYaml object at 0x7fb45d586850>
Initialize params from:./module/shufflenetv2.pth
----------------------------------------------------------------
        Layer (type)               Output Shape         Param #
================================================================
            Conv2d-1         [-1, 24, 176, 176]             648
       BatchNorm2d-2         [-1, 24, 176, 176]              48
              ReLU-3         [-1, 24, 176, 176]               0
         MaxPool2d-4           [-1, 24, 88, 88]               0
            Conv2d-5           [-1, 24, 44, 44]             216
       BatchNorm2d-6           [-1, 24, 44, 44]              48
            Conv2d-7           [-1, 24, 44, 44]             576
       BatchNorm2d-8           [-1, 24, 44, 44]              48
              ReLU-9           [-1, 24, 44, 44]               0
           Conv2d-10           [-1, 24, 88, 88]             576
      BatchNorm2d-11           [-1, 24, 88, 88]              48
             ReLU-12           [-1, 24, 88, 88]               0
           Conv2d-13           [-1, 24, 44, 44]             216
      BatchNorm2d-14           [-1, 24, 44, 44]              48
           Conv2d-15           [-1, 24, 44, 44]             576
      BatchNorm2d-16           [-1, 24, 44, 44]              48
             ReLU-17           [-1, 24, 44, 44]               0
   ShuffleV2Block-18           [-1, 48, 44, 44]               0
           Conv2d-19           [-1, 24, 44, 44]             576
      BatchNorm2d-20           [-1, 24, 44, 44]              48
             ReLU-21           [-1, 24, 44, 44]               0
           Conv2d-22           [-1, 24, 44, 44]             216
      BatchNorm2d-23           [-1, 24, 44, 44]              48
           Conv2d-24           [-1, 24, 44, 44]             576
      BatchNorm2d-25           [-1, 24, 44, 44]              48
             ReLU-26           [-1, 24, 44, 44]               0
   ShuffleV2Block-27           [-1, 48, 44, 44]               0
           Conv2d-28           [-1, 24, 44, 44]             576
      BatchNorm2d-29           [-1, 24, 44, 44]              48
             ReLU-30           [-1, 24, 44, 44]               0
           Conv2d-31           [-1, 24, 44, 44]             216
      BatchNorm2d-32           [-1, 24, 44, 44]              48
           Conv2d-33           [-1, 24, 44, 44]             576
      BatchNorm2d-34           [-1, 24, 44, 44]              48
             ReLU-35           [-1, 24, 44, 44]               0
   ShuffleV2Block-36           [-1, 48, 44, 44]               0
           Conv2d-37           [-1, 24, 44, 44]             576
      BatchNorm2d-38           [-1, 24, 44, 44]              48
             ReLU-39           [-1, 24, 44, 44]               0
           Conv2d-40           [-1, 24, 44, 44]             216
      BatchNorm2d-41           [-1, 24, 44, 44]              48
           Conv2d-42           [-1, 24, 44, 44]             576
      BatchNorm2d-43           [-1, 24, 44, 44]              48
             ReLU-44           [-1, 24, 44, 44]               0
   ShuffleV2Block-45           [-1, 48, 44, 44]               0
           Conv2d-46           [-1, 48, 22, 22]             432
      BatchNorm2d-47           [-1, 48, 22, 22]              96
           Conv2d-48           [-1, 48, 22, 22]           2,304
      BatchNorm2d-49           [-1, 48, 22, 22]              96
             ReLU-50           [-1, 48, 22, 22]               0
           Conv2d-51           [-1, 48, 44, 44]           2,304
      BatchNorm2d-52           [-1, 48, 44, 44]              96
             ReLU-53           [-1, 48, 44, 44]               0
           Conv2d-54           [-1, 48, 22, 22]             432
      BatchNorm2d-55           [-1, 48, 22, 22]              96
           Conv2d-56           [-1, 48, 22, 22]           2,304
      BatchNorm2d-57           [-1, 48, 22, 22]              96
             ReLU-58           [-1, 48, 22, 22]               0
   ShuffleV2Block-59           [-1, 96, 22, 22]               0
           Conv2d-60           [-1, 48, 22, 22]           2,304
      BatchNorm2d-61           [-1, 48, 22, 22]              96
             ReLU-62           [-1, 48, 22, 22]               0
           Conv2d-63           [-1, 48, 22, 22]             432
      BatchNorm2d-64           [-1, 48, 22, 22]              96
           Conv2d-65           [-1, 48, 22, 22]           2,304
      BatchNorm2d-66           [-1, 48, 22, 22]              96
             ReLU-67           [-1, 48, 22, 22]               0
   ShuffleV2Block-68           [-1, 96, 22, 22]               0
           Conv2d-69           [-1, 48, 22, 22]           2,304
      BatchNorm2d-70           [-1, 48, 22, 22]              96
             ReLU-71           [-1, 48, 22, 22]               0
           Conv2d-72           [-1, 48, 22, 22]             432
      BatchNorm2d-73           [-1, 48, 22, 22]              96
           Conv2d-74           [-1, 48, 22, 22]           2,304
      BatchNorm2d-75           [-1, 48, 22, 22]              96
             ReLU-76           [-1, 48, 22, 22]               0
   ShuffleV2Block-77           [-1, 96, 22, 22]               0
           Conv2d-78           [-1, 48, 22, 22]           2,304
      BatchNorm2d-79           [-1, 48, 22, 22]              96
             ReLU-80           [-1, 48, 22, 22]               0
           Conv2d-81           [-1, 48, 22, 22]             432
      BatchNorm2d-82           [-1, 48, 22, 22]              96
           Conv2d-83           [-1, 48, 22, 22]           2,304
      BatchNorm2d-84           [-1, 48, 22, 22]              96
             ReLU-85           [-1, 48, 22, 22]               0
   ShuffleV2Block-86           [-1, 96, 22, 22]               0
           Conv2d-87           [-1, 48, 22, 22]           2,304
      BatchNorm2d-88           [-1, 48, 22, 22]              96
             ReLU-89           [-1, 48, 22, 22]               0
           Conv2d-90           [-1, 48, 22, 22]             432
      BatchNorm2d-91           [-1, 48, 22, 22]              96
           Conv2d-92           [-1, 48, 22, 22]           2,304
      BatchNorm2d-93           [-1, 48, 22, 22]              96
             ReLU-94           [-1, 48, 22, 22]               0
   ShuffleV2Block-95           [-1, 96, 22, 22]               0
           Conv2d-96           [-1, 48, 22, 22]           2,304
      BatchNorm2d-97           [-1, 48, 22, 22]              96
             ReLU-98           [-1, 48, 22, 22]               0
           Conv2d-99           [-1, 48, 22, 22]             432
     BatchNorm2d-100           [-1, 48, 22, 22]              96
          Conv2d-101           [-1, 48, 22, 22]           2,304
     BatchNorm2d-102           [-1, 48, 22, 22]              96
            ReLU-103           [-1, 48, 22, 22]               0
  ShuffleV2Block-104           [-1, 96, 22, 22]               0
          Conv2d-105           [-1, 48, 22, 22]           2,304
     BatchNorm2d-106           [-1, 48, 22, 22]              96
            ReLU-107           [-1, 48, 22, 22]               0
          Conv2d-108           [-1, 48, 22, 22]             432
     BatchNorm2d-109           [-1, 48, 22, 22]              96
          Conv2d-110           [-1, 48, 22, 22]           2,304
     BatchNorm2d-111           [-1, 48, 22, 22]              96
            ReLU-112           [-1, 48, 22, 22]               0
  ShuffleV2Block-113           [-1, 96, 22, 22]               0
          Conv2d-114           [-1, 48, 22, 22]           2,304
     BatchNorm2d-115           [-1, 48, 22, 22]              96
            ReLU-116           [-1, 48, 22, 22]               0
          Conv2d-117           [-1, 48, 22, 22]             432
     BatchNorm2d-118           [-1, 48, 22, 22]              96
          Conv2d-119           [-1, 48, 22, 22]           2,304
     BatchNorm2d-120           [-1, 48, 22, 22]              96
            ReLU-121           [-1, 48, 22, 22]               0
  ShuffleV2Block-122           [-1, 96, 22, 22]               0
          Conv2d-123           [-1, 96, 11, 11]             864
     BatchNorm2d-124           [-1, 96, 11, 11]             192
          Conv2d-125           [-1, 96, 11, 11]           9,216
     BatchNorm2d-126           [-1, 96, 11, 11]             192
            ReLU-127           [-1, 96, 11, 11]               0
          Conv2d-128           [-1, 96, 22, 22]           9,216
     BatchNorm2d-129           [-1, 96, 22, 22]             192
            ReLU-130           [-1, 96, 22, 22]               0
          Conv2d-131           [-1, 96, 11, 11]             864
     BatchNorm2d-132           [-1, 96, 11, 11]             192
          Conv2d-133           [-1, 96, 11, 11]           9,216
     BatchNorm2d-134           [-1, 96, 11, 11]             192
            ReLU-135           [-1, 96, 11, 11]               0
  ShuffleV2Block-136          [-1, 192, 11, 11]               0
          Conv2d-137           [-1, 96, 11, 11]           9,216
     BatchNorm2d-138           [-1, 96, 11, 11]             192
            ReLU-139           [-1, 96, 11, 11]               0
          Conv2d-140           [-1, 96, 11, 11]             864
     BatchNorm2d-141           [-1, 96, 11, 11]             192
          Conv2d-142           [-1, 96, 11, 11]           9,216
     BatchNorm2d-143           [-1, 96, 11, 11]             192
            ReLU-144           [-1, 96, 11, 11]               0
  ShuffleV2Block-145          [-1, 192, 11, 11]               0
          Conv2d-146           [-1, 96, 11, 11]           9,216
     BatchNorm2d-147           [-1, 96, 11, 11]             192
            ReLU-148           [-1, 96, 11, 11]               0
          Conv2d-149           [-1, 96, 11, 11]             864
     BatchNorm2d-150           [-1, 96, 11, 11]             192
          Conv2d-151           [-1, 96, 11, 11]           9,216
     BatchNorm2d-152           [-1, 96, 11, 11]             192
            ReLU-153           [-1, 96, 11, 11]               0
  ShuffleV2Block-154          [-1, 192, 11, 11]               0
          Conv2d-155           [-1, 96, 11, 11]           9,216
     BatchNorm2d-156           [-1, 96, 11, 11]             192
            ReLU-157           [-1, 96, 11, 11]               0
          Conv2d-158           [-1, 96, 11, 11]             864
     BatchNorm2d-159           [-1, 96, 11, 11]             192
          Conv2d-160           [-1, 96, 11, 11]           9,216
     BatchNorm2d-161           [-1, 96, 11, 11]             192
            ReLU-162           [-1, 96, 11, 11]               0
  ShuffleV2Block-163          [-1, 192, 11, 11]               0
    ShuffleNetV2-164  [[-1, 48, 44, 44], [-1, 96, 22, 22], [-1, 192, 11, 11]]               0
        Upsample-165          [-1, 192, 22, 22]               0
       AvgPool2d-166           [-1, 48, 22, 22]               0
          Conv2d-167           [-1, 96, 22, 22]          32,256
     BatchNorm2d-168           [-1, 96, 22, 22]             192
            ReLU-169           [-1, 96, 22, 22]               0
         Conv1x1-170           [-1, 96, 22, 22]               0
          Conv2d-171           [-1, 96, 22, 22]           2,400
     BatchNorm2d-172           [-1, 96, 22, 22]             192
            ReLU-173           [-1, 96, 22, 22]               0
          Conv2d-174           [-1, 96, 22, 22]           2,400
     BatchNorm2d-175           [-1, 96, 22, 22]             192
            ReLU-176           [-1, 96, 22, 22]               0
          Conv2d-177           [-1, 96, 22, 22]           2,400
     BatchNorm2d-178           [-1, 96, 22, 22]             192
            ReLU-179           [-1, 96, 22, 22]               0
          Conv2d-180           [-1, 96, 22, 22]           2,400
     BatchNorm2d-181           [-1, 96, 22, 22]             192
            ReLU-182           [-1, 96, 22, 22]               0
          Conv2d-183           [-1, 96, 22, 22]           2,400
     BatchNorm2d-184           [-1, 96, 22, 22]             192
            ReLU-185           [-1, 96, 22, 22]               0
          Conv2d-186           [-1, 96, 22, 22]           2,400
     BatchNorm2d-187           [-1, 96, 22, 22]             192
            ReLU-188           [-1, 96, 22, 22]               0
          Conv2d-189           [-1, 96, 22, 22]          27,648
     BatchNorm2d-190           [-1, 96, 22, 22]             192
            ReLU-191           [-1, 96, 22, 22]               0
             SPP-192           [-1, 96, 22, 22]               0
          Conv2d-193           [-1, 96, 22, 22]           9,216
     BatchNorm2d-194           [-1, 96, 22, 22]             192
            ReLU-195           [-1, 96, 22, 22]               0
         Conv1x1-196           [-1, 96, 22, 22]               0
          Conv2d-197           [-1, 96, 22, 22]           2,400
     BatchNorm2d-198           [-1, 96, 22, 22]             192
            ReLU-199           [-1, 96, 22, 22]               0
          Conv2d-200            [-1, 1, 22, 22]              96
     BatchNorm2d-201            [-1, 1, 22, 22]               2
            Head-202            [-1, 1, 22, 22]               0
         Sigmoid-203            [-1, 1, 22, 22]               0
          Conv2d-204           [-1, 96, 22, 22]           2,400
     BatchNorm2d-205           [-1, 96, 22, 22]             192
            ReLU-206           [-1, 96, 22, 22]               0
          Conv2d-207            [-1, 4, 22, 22]             384
     BatchNorm2d-208            [-1, 4, 22, 22]               8
            Head-209            [-1, 4, 22, 22]               0
          Conv2d-210           [-1, 96, 22, 22]           2,400
     BatchNorm2d-211           [-1, 96, 22, 22]             192
            ReLU-212           [-1, 96, 22, 22]               0
          Conv2d-213            [-1, 2, 22, 22]             192
     BatchNorm2d-214            [-1, 2, 22, 22]               4
            Head-215            [-1, 2, 22, 22]               0
         Softmax-216            [-1, 2, 22, 22]               0
      DetectHead-217            [-1, 7, 22, 22]               0
================================================================
Total params: 236,846
Trainable params: 236,846
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 1.42
Forward/backward pass size (MB): 765314425.85
Params size (MB): 0.90
Estimated Total Size (MB): 765314428.17
----------------------------------------------------------------
use SGD optimizer
Starting training for 10 epochs...
Epoch:0 LR:0.000001 IOU:0.525141 Obj:0.114225 Cls:0.910879 Total:6.939609: 100%|███████████████████████████████████████████████████████████████████████████████████████████| 16/16 [00:03<00:00,  5.21it/s]
Epoch:1 LR:0.000023 IOU:0.523991 Obj:0.114126 Cls:0.805686 Total:6.823635: 100%|███████████████████████████████████████████████████████████████████████████████████████████| 16/16 [00:02<00:00,  5.76it/s]
Epoch:2 LR:0.000119 IOU:0.487222 Obj:0.112220 Cls:0.629635 Total:6.322935: 100%|███████████████████████████████████████████████████████████████████████████████████████████| 16/16 [00:02<00:00,  5.64it/s]
Epoch:3 LR:0.000385 IOU:0.431137 Obj:0.106160 Cls:0.331946 Total:5.479599: 100%|███████████████████████████████████████████████████████████████████████████████████████████| 16/16 [00:02<00:00,  5.78it/s]
Epoch:4 LR:0.000951 IOU:0.370437 Obj:0.081730 Cls:0.194502 Total:4.465681: 100%|███████████████████████████████████████████████████████████████████████████████████████████| 16/16 [00:02<00:00,  5.65it/s]
Epoch:5 LR:0.001000 IOU:0.401281 Obj:0.054203 Cls:0.165006 Total:4.242496: 100%|███████████████████████████████████████████████████████████████████████████████████████████| 16/16 [00:02<00:00,  5.63it/s]
Epoch:6 LR:0.001000 IOU:0.341546 Obj:0.043842 Cls:0.091943 Total:3.525779: 100%|███████████████████████████████████████████████████████████████████████████████████████████| 16/16 [00:02<00:00,  5.68it/s]
Epoch:7 LR:0.001000 IOU:0.334242 Obj:0.038874 Cls:0.078698 Total:3.374617: 100%|███████████████████████████████████████████████████████████████████████████████████████████| 16/16 [00:02<00:00,  5.68it/s]
Epoch:8 LR:0.001000 IOU:0.319824 Obj:0.035644 Cls:0.120743 Total:3.249641: 100%|███████████████████████████████████████████████████████████████████████████████████████████| 16/16 [00:02<00:00,  5.87it/s]
Epoch:9 LR:0.001000 IOU:0.306538 Obj:0.034842 Cls:0.099246 Total:3.109022: 100%|███████████████████████████████████████████████████████████████████████████████████████████| 16/16 [00:02<00:00,  5.74it/s]
Epoch:10 LR:0.001000 IOU:0.299839 Obj:0.032444 Cls:0.059097 Total:2.976912: 100%|██████████████████████████████████████████████████████████████████████████████████████████| 16/16 [00:02<00:00,  5.83it/s]
computer mAP...
  0%|                                                                                                                                                                                | 0/3 [00:00<?, ?it/s]/home/kurisaw/anaconda3/envs/yolo-fastesdet/lib/python3.8/site-packages/torch/functional.py:568: UserWarning: torch.meshgrid: in an upcoming release, it will be required to pass the indexing argument. (Triggered internally at  /opt/conda/conda-bld/pytorch_1646755903507/work/aten/src/ATen/native/TensorShape.cpp:2228.)
  return _VF.meshgrid(tensors, **kwargs)  # type: ignore[attr-defined]
100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 3/3 [00:01<00:00,  1.72it/s]
creating index...
index created!
creating index...
index created!
Running per image evaluation...
Evaluate annotation type *bbox*
DONE (t=0.31s).
Accumulating evaluation results...
DONE (t=0.11s).
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.025
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.126
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.001
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.000
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.021
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.037
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.048
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.143
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.152
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.000
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.080
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.226

[kurisaW]

分析

根据给出的日志，可以分析训练结果如下：

• 训练过程中使用了3个批次（batches）进行计算。
• 训练过程中的平均精确度（Average Precision，AP）和召回率（Average Recall，AR）的评估结果如下：
  • 在所有IoU（Intersection over Union）阈值范围内，所有尺度（area）和最大检测数（maxDets）下的平均精确度（AP）为0.025。这表示在不同IoU阈值下综合考虑所有尺度和最大检测数的情况下，模型的平均精确度较低。
  • 在IoU阈值为0.50的情况下，所有尺度和最大检测数下的平均精确度（AP）为0.126。这表示在相对较宽松的IoU阈值下，模型的平均精确度较高。
  • 在IoU阈值为0.75的情况下，所有尺度和最大检测数下的平均精确度（AP）为0.001。这表示在较为严格的IoU阈值下，模型的平均精确度较低。
  • 在小尺度物体上（area=small），所有IoU阈值范围内的平均精确度（AP）为0.000。这表示对于小尺度物体，模型的平均精确度较低。
  • 在中等尺度物体上（area=medium），所有IoU阈值范围内的平均精确度（AP）为0.021。这表示对于中等尺度物体，模型的平均精确度较低。
  • 在大尺度物体上（area=large），所有IoU阈值范围内的平均精确度（AP）为0.037。这表示对于大尺度物体，模型的平均精确度较低。
  • 在所有IoU阈值范围内，所有尺度和最大检测数下的平均召回率（AR）为0.152。这表示在不同IoU阈值下综合考虑所有尺度和最大检测数的情况下，模型的平均召回率较低。
  • 在小尺度物体上（area=small），所有IoU阈值范围内的平均召回率（AR）为0.000。这表示对于小尺度物体，模型的平均召回率较低。
  • 在中等尺度物体上（area=medium），所有IoU阈值范围内的平均召回率（AR）为0.080。这表示对于中等尺度物体，模型的平均召回率较低。
  • 在大尺度物体上（area=large），所有IoU阈值范围内的平均召回率（AR）为0.226。这表示对于大尺度物体，模型的平均召回率较低。

综合来看，该模型在综合考虑所有尺度和最大检测数的情况下的平均精确度和召回率都较低。特别是对于小尺度物体和较高IoU阈值的情况下，模型表现不佳。为了提高模型性能，可以尝试优化模型架构、调整训练参数、增加训练数据量或进行数据增强等方法。同时，根据实际需求，还可以根据不同尺度和IoU阈值对模型进行评估和优化。