YOLO v5 recognition

Stuff I am working on
Daniel Wee
Site Admin
Posts: 2245
Joined: Wed 25 Feb 25 2009 8:00 pm

Re: YoloV5 recognition

Post by Daniel Wee » Fri 04 Sep 04 2020 8:11 pm

https://github.com/ultralytics/yolov5/w ... ustom-Data

It seems that if the training succeeds, this should be reflected in the yolov5/runs/expnn/test_batch0_pred .jpg image. At this point, all the _pred output images do not show any bounding boxes that indicate successful detection. This only applies to the output from the first epoch though, and since we are training with --weights '' one would imagine that nothing will match after just one epoch?

As we continue to investigate the training failure, I am now looking at a more in-depth explanation of the dataset preparation process:-

https://towardsdatascience.com/training ... 81fa480af0

In particular, the following steps are mentioned:-
  • Verifying your annotations are correct (e.g. none of the annotations are out of frame in the images)
  • Ensuring the EXIF orientation of your images is correct (i.e. your images are stored on disk differently than how you view them in applications, see more)
  • Resizing images and updating image annotations to match the newly sized images
  • Various color corrections that may improve model performance like grayscale and contrast adjustments
  • Formatting annotations to match the requirements of your model’s inputs (e.g. generating TFRecords for TensorFlow or a flat text file for some implementations of YOLO).
On top of HyperLabel (Windows), we installed labelImg on Ubuntu as a backup checker. This can be pip3 installed according to:-

https://github.com/tzutalin/labelImg

One should also create a text file called yolov5/training/predefined.txt that contains just one line: "lizard"

The command should be:-

Code: Select all

sudo -H pip3 install labelImg
It is then executed in the yolov5/training directory from the terminal commandline:-

Code: Select all

labelImg img predefined.txt
Daniel Wee
Site Admin
Posts: 2245
Joined: Wed 25 Feb 25 2009 8:00 pm

Re: YoloV5 recognition

Post by Daniel Wee » Fri 04 Sep 04 2020 8:39 pm

At this point we are resizing all the images to a width of 640 using Corel PSP's batch resizer. This is then checked with labelImg to ensure all the bounding boxes are correct. This needs to be re-split using the train_val_folder_split.py script. This should be done after clearing out the output directories, including the label and cache files. This is found in the data/ directory and its subdirectories.

Once that is done, we will re-train using the original train command:-

Code: Select all

python train.py --img 640 --batch 16 --epochs 300 --data training/dataset.yaml --cfg training/yolov5l.yaml --weights ''
At this point, our training set consists of 233-images and the validation set consists of 57-images.

The training epoch output indicates the 57-images with 70-targets. It is unclear what the targets refer to but possibly where there are multiple objects per image.

Result

Having done the training with a verified dataset of size 640 (width), we are still NOT getting detection working. I think we have to assume that some other problem is occurring at the training level or that there simply is not enough training samples for the training to be successful. At this point, the strategy going forward would be to try on a known working setup example to see if we are getting any results. This is to verify that the YOLO framework is working in the first place. Secondly it would be to investigate if the number of samples in the training set are sufficient.
Daniel Wee
Site Admin
Posts: 2245
Joined: Wed 25 Feb 25 2009 8:00 pm

Re: YoloV5 recognition

Post by Daniel Wee » Sat 05 Sep 05 2020 11:48 am

So I followed the example for training COCO128 from:-

https://github.com/ultralytics/yolov5/w ... ustom-Data

Unfortunately I was getting no results here either, indicating that there might be a configuration problem in my yolov5 setup or in the CUDA or cuDNN setup. Will investigate this further.

One way of verifying the issue is to try running this project in Google Colab:-

https://colab.research.google.com/notebooks/

This involves creating a new Notebook. We will copy our yolov5 directory across into the Colab directory in MyDrive. This is only accessible through the browser so I'm copying this into the local Google Drive directory and them using the browser interface to move it into the Colab directory.
Daniel Wee
Site Admin
Posts: 2245
Joined: Wed 25 Feb 25 2009 8:00 pm

Re: YoloV5 recognition

Post by Daniel Wee » Sat 05 Sep 05 2020 1:55 pm

We used Roboflow to produce our training data and uploaded the resultant file to:-

http://www.tsebi.com/reviews/Lizards.v1 ... ytorch.zip

This was then used in the Roboflow Colab script:-

https://colab.research.google.com/drive ... KEqYtTgbRc

The training result was:-

Code: Select all

/content/yolov5
Using CUDA device0 _CudaDeviceProperties(name='Tesla P100-PCIE-16GB', total_memory=16280MB)

Namespace(adam=False, batch_size=16, bucket='', cache_images=True, cfg='./models/custom_yolov5s.yaml', data='../data.yaml', device='', epochs=100, evolve=False, global_rank=-1, hyp='data/hyp.scratch.yaml', image_weights=False, img_size=[416, 416], local_rank=-1, logdir='runs/', multi_scale=False, name='yolov5s_results', noautoanchor=False, nosave=False, notest=False, rect=False, resume=False, single_cls=False, sync_bn=False, total_batch_size=16, weights='', workers=8, world_size=1)
Start Tensorboard with "tensorboard --logdir runs/", view at http://localhost:6006/
2020-09-05 05:40:27.383932: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcudart.so.10.1
Hyperparameters {'lr0': 0.01, 'lrf': 0.2, 'momentum': 0.937, 'weight_decay': 0.0005, 'giou': 0.05, 'cls': 0.5, 'cls_pw': 1.0, 'obj': 1.0, 'obj_pw': 1.0, 'iou_t': 0.2, 'anchor_t': 4.0, 'fl_gamma': 0.0, 'hsv_h': 0.015, 'hsv_s': 0.7, 'hsv_v': 0.4, 'degrees': 0.0, 'translate': 0.1, 'scale': 0.5, 'shear': 0.0, 'perspective': 0.0, 'flipud': 0.0, 'fliplr': 0.5, 'mixup': 0.0}

                 from  n    params  module                                  arguments                     
  0                -1  1      3520  models.common.Focus                     [3, 32, 3]                    
  1                -1  1     18560  models.common.Conv                      [32, 64, 3, 2]                
  2                -1  1     19904  models.common.BottleneckCSP             [64, 64, 1]                   
  3                -1  1     73984  models.common.Conv                      [64, 128, 3, 2]               
  4                -1  1    161152  models.common.BottleneckCSP             [128, 128, 3]                 
  5                -1  1    295424  models.common.Conv                      [128, 256, 3, 2]              
  6                -1  1    641792  models.common.BottleneckCSP             [256, 256, 3]                 
  7                -1  1   1180672  models.common.Conv                      [256, 512, 3, 2]              
  8                -1  1    656896  models.common.SPP                       [512, 512, [5, 9, 13]]        
  9                -1  1   1248768  models.common.BottleneckCSP             [512, 512, 1, False]          
 10                -1  1    131584  models.common.Conv                      [512, 256, 1, 1]              
 11                -1  1         0  torch.nn.modules.upsampling.Upsample    [None, 2, 'nearest']          
 12           [-1, 6]  1         0  models.common.Concat                    [1]                           
 13                -1  1    378624  models.common.BottleneckCSP             [512, 256, 1, False]          
 14                -1  1     33024  models.common.Conv                      [256, 128, 1, 1]              
 15                -1  1         0  torch.nn.modules.upsampling.Upsample    [None, 2, 'nearest']          
 16           [-1, 4]  1         0  models.common.Concat                    [1]                           
 17                -1  1     95104  models.common.BottleneckCSP             [256, 128, 1, False]          
 18                -1  1    147712  models.common.Conv                      [128, 128, 3, 2]              
 19          [-1, 14]  1         0  models.common.Concat                    [1]                           
 20                -1  1    313088  models.common.BottleneckCSP             [256, 256, 1, False]          
 21                -1  1    590336  models.common.Conv                      [256, 256, 3, 2]              
 22          [-1, 10]  1         0  models.common.Concat                    [1]                           
 23                -1  1   1248768  models.common.BottleneckCSP             [512, 512, 1, False]          
 24      [17, 20, 23]  1     16182  models.yolo.Detect                      [1, [[10, 13, 16, 30, 33, 23], [30, 61, 62, 45, 59, 119], [116, 90, 156, 198, 373, 326]], [128, 256, 512]]
Model Summary: 191 layers, 7.25509e+06 parameters, 7.25509e+06 gradients

Optimizer groups: 62 .bias, 70 conv.weight, 59 other
Scanning images: 100% 185/185 [00:00<00:00, 3439.98it/s]
Scanning labels ../train/labels.cache (185 found, 0 missing, 0 empty, 0 duplicate, for 185 images): 185it [00:00, 13115.18it/s]
Caching images (0.1GB): 100% 185/185 [00:00<00:00, 393.90it/s]
Scanning images: 100% 53/53 [00:00<00:00, 3264.43it/s]
Scanning labels ../valid/labels.cache (53 found, 0 missing, 0 empty, 0 duplicate, for 53 images): 53it [00:00, 9986.44it/s]
Caching images (0.0GB): 100% 53/53 [00:00<00:00, 173.38it/s]
/usr/local/lib/python3.6/dist-packages/statsmodels/tools/_testing.py:19: FutureWarning: pandas.util.testing is deprecated. Use the functions in the public API at pandas.testing instead.
  import pandas.util.testing as tm

Analyzing anchors... anchors/target = 3.45, Best Possible Recall (BPR) = 1.0000
Image sizes 416 train, 416 test
Using 2 dataloader workers
Logging results to runs/exp0_yolov5s_results
Starting training for 100 epochs...

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
      0/99    0.818G    0.1114    0.0483         0    0.1597        23       416: 100% 12/12 [00:05<00:00,  2.16it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:03<00:00,  1.07it/s]
                 all          53          54           0           0     0.00436    0.000588

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
      1/99     1.42G   0.09849   0.05125         0    0.1497        20       416: 100% 12/12 [00:02<00:00,  4.30it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.34it/s]
                 all          53          54           0           0    0.000416    4.45e-05

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
      2/99     1.42G   0.08871   0.05231         0     0.141        25       416: 100% 12/12 [00:02<00:00,  5.02it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.55it/s]
                 all          53          54           0           0    0.000529    5.74e-05

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
      3/99     1.42G   0.08284    0.0574         0    0.1402        24       416: 100% 12/12 [00:02<00:00,  4.94it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.57it/s]
                 all          53          54           0           0    0.000999      0.0001

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
      4/99     1.42G   0.08285   0.05376         0    0.1366        29       416: 100% 12/12 [00:02<00:00,  4.94it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.43it/s]
                 all          53          54           0           0    0.000634    6.34e-05

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
      5/99     1.42G   0.07538   0.05296         0    0.1283        21       416: 100% 12/12 [00:02<00:00,  4.93it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.58it/s]
                 all          53          54           0           0     0.00087    0.000102

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
      6/99     1.42G   0.07737   0.05539         0    0.1328        22       416: 100% 12/12 [00:02<00:00,  4.93it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.56it/s]
                 all          53          54           0           0    0.000336     5.7e-05

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
      7/99     1.42G   0.07874   0.05205         0    0.1308        18       416: 100% 12/12 [00:02<00:00,  4.85it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.52it/s]
                 all          53          54           0           0    0.000897    0.000138

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
      8/99     1.42G   0.07445   0.05737         0    0.1318        22       416: 100% 12/12 [00:02<00:00,  4.93it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.37it/s]
                 all          53          54           0           0    0.000596    7.01e-05

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
      9/99     1.42G   0.07856   0.06159         0    0.1401        25       416: 100% 12/12 [00:02<00:00,  4.92it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.55it/s]
                 all          53          54           0           0    0.000892    0.000117

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     10/99     1.42G   0.07614   0.05583         0     0.132        21       416: 100% 12/12 [00:02<00:00,  4.94it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.52it/s]
                 all          53          54           0           0    0.000844    9.66e-05

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     11/99     1.42G   0.07376   0.05572         0    0.1295        17       416: 100% 12/12 [00:02<00:00,  4.89it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.39it/s]
                 all          53          54           0           0     0.00837    0.000894

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     12/99     1.42G   0.07735   0.05711         0    0.1345        25       416: 100% 12/12 [00:02<00:00,  4.85it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.52it/s]
                 all          53          54           0           0     0.00651     0.00132

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     13/99     1.42G   0.07779   0.05021         0     0.128        20       416: 100% 12/12 [00:02<00:00,  4.98it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.53it/s]
                 all          53          54           0           0     0.00293     0.00045

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     14/99     1.42G   0.07604   0.05829         0    0.1343        29       416: 100% 12/12 [00:02<00:00,  5.41it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.31it/s]
                 all          53          54           0           0     0.00275    0.000282

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     15/99     1.42G   0.07381   0.06115         0     0.135        18       416: 100% 12/12 [00:02<00:00,  5.57it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.53it/s]
                 all          53          54       0.116      0.0185      0.0135     0.00143

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     16/99     1.42G    0.0745   0.05586         0    0.1304        24       416: 100% 12/12 [00:02<00:00,  5.43it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.49it/s]
                 all          53          54       0.242      0.0185      0.0107     0.00164

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     17/99     1.42G   0.07581   0.05233         0    0.1281        19       416: 100% 12/12 [00:02<00:00,  5.55it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.43it/s]
                 all          53          54       0.154      0.0185      0.0186     0.00205

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     18/99     1.42G   0.07427   0.05633         0    0.1306        22       416: 100% 12/12 [00:02<00:00,  5.49it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.53it/s]
                 all          53          54           0           0     0.00197    0.000257

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     19/99     1.42G   0.07496   0.05468         0    0.1296        23       416: 100% 12/12 [00:02<00:00,  5.64it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.53it/s]
                 all          53          54           0           0     0.00312    0.000418

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     20/99     1.42G   0.07526   0.05861         0    0.1339        28       416: 100% 12/12 [00:02<00:00,  5.45it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.48it/s]
                 all          53          54           0           0     0.00424      0.0011

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     21/99     1.42G   0.07881   0.05913         0    0.1379        23       416: 100% 12/12 [00:02<00:00,  5.58it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.56it/s]
                 all          53          54           0           0     0.00906     0.00272

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     22/99     1.42G   0.07463   0.05595         0    0.1306        20       416: 100% 12/12 [00:02<00:00,  5.44it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.50it/s]
                 all          53          54           0           0      0.0116     0.00207

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     23/99     1.42G   0.07444   0.05956         0     0.134        24       416: 100% 12/12 [00:02<00:00,  5.45it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.37it/s]
                 all          53          54       0.206      0.0153      0.0181     0.00462

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     24/99     1.42G   0.07505   0.06286         0    0.1379        22       416: 100% 12/12 [00:02<00:00,  5.50it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.55it/s]
                 all          53          54       0.217     0.00803      0.0244     0.00308

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     25/99     1.42G   0.07804   0.05638         0    0.1344        23       416: 100% 12/12 [00:02<00:00,  5.61it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.54it/s]
                 all          53          54       0.289      0.0161      0.0208     0.00272

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     26/99     1.42G   0.07333   0.06168         0     0.135        29       416: 100% 12/12 [00:02<00:00,  5.50it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.39it/s]
                 all          53          54      0.0879      0.0185      0.0263     0.00613

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     27/99     1.42G   0.07237   0.06148         0    0.1339        21       416: 100% 12/12 [00:02<00:00,  5.48it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.43it/s]
                 all          53          54           0           0      0.0218     0.00335

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     28/99     1.42G   0.07335   0.06151         0    0.1349        24       416: 100% 12/12 [00:02<00:00,  5.44it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.50it/s]
                 all          53          54           0           0      0.0226     0.00303

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     29/99     1.42G   0.07314    0.0578         0    0.1309        20       416: 100% 12/12 [00:02<00:00,  5.58it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.41it/s]
                 all          53          54           0           0      0.0232     0.00317

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     30/99     1.42G   0.06684   0.05659         0    0.1234        15       416: 100% 12/12 [00:02<00:00,  5.54it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.54it/s]
                 all          53          54           0           0      0.0308     0.00476

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     31/99     1.42G   0.06864   0.05874         0    0.1274        23       416: 100% 12/12 [00:02<00:00,  5.48it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.56it/s]
                 all          53          54           0           0      0.0244     0.00327

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     32/99     1.42G   0.07183   0.06147         0    0.1333        23       416: 100% 12/12 [00:02<00:00,  5.59it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.41it/s]
                 all          53          54           0           0      0.0173     0.00286

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     33/99     1.42G   0.07031   0.05899         0    0.1293        23       416: 100% 12/12 [00:02<00:00,  5.63it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.58it/s]
                 all          53          54           0           0      0.0111     0.00151

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     34/99     1.42G    0.0705   0.05949         0      0.13        19       416: 100% 12/12 [00:02<00:00,  5.63it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.53it/s]
                 all          53          54           0           0      0.0123     0.00144

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     35/99     1.42G   0.07197   0.06245         0    0.1344        27       416: 100% 12/12 [00:02<00:00,  5.52it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.36it/s]
                 all          53          54           0           0      0.0129     0.00174

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     36/99     1.42G   0.07251   0.05924         0    0.1317        22       416: 100% 12/12 [00:02<00:00,  5.56it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.52it/s]
                 all          53          54           0           0      0.0213     0.00288

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     37/99     1.42G   0.07174     0.062         0    0.1337        30       416: 100% 12/12 [00:02<00:00,  5.57it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.54it/s]
                 all          53          54           0           0      0.0381     0.00602

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     38/99     1.42G   0.07479   0.05834         0    0.1331        16       416: 100% 12/12 [00:02<00:00,  5.46it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.33it/s]
                 all          53          54           0           0       0.026     0.00329

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     39/99     1.42G   0.07053   0.05913         0    0.1297        27       416: 100% 12/12 [00:02<00:00,  5.47it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.56it/s]
                 all          53          54           1      0.0185      0.0338     0.00367

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     40/99     1.42G   0.07067   0.05806         0    0.1287        21       416: 100% 12/12 [00:02<00:00,  5.57it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.47it/s]
                 all          53          54           1      0.0185      0.0666      0.0125

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     41/99     1.42G   0.06839   0.06284         0    0.1312        20       416: 100% 12/12 [00:02<00:00,  5.52it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.59it/s]
                 all          53          54           0           0      0.0416     0.00704

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     42/99     1.42G   0.06843    0.0616         0      0.13        18       416: 100% 12/12 [00:02<00:00,  5.75it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.37it/s]
                 all          53          54           0           0      0.0261     0.00457

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     43/99     1.42G   0.06933   0.06512         0    0.1345        26       416: 100% 12/12 [00:02<00:00,  5.62it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.52it/s]
                 all          53          54           0           0      0.0244     0.00377

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     44/99     1.42G    0.0676   0.06698         0    0.1346        26       416: 100% 12/12 [00:02<00:00,  5.73it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.55it/s]
                 all          53          54           0           0        0.02     0.00243

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     45/99     1.42G   0.06492   0.06747         0    0.1324        36       416: 100% 12/12 [00:02<00:00,  5.74it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.41it/s]
                 all          53          54           0           0      0.0209     0.00274

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     46/99     1.42G   0.06411   0.06405         0    0.1282        24       416: 100% 12/12 [00:02<00:00,  5.74it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.58it/s]
                 all          53          54           0           0      0.0241     0.00279

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     47/99     1.42G   0.06485   0.05992         0    0.1248        23       416: 100% 12/12 [00:02<00:00,  5.73it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.51it/s]
                 all          53          54           0           0      0.0142     0.00214

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     48/99     1.42G   0.06325   0.06262         0    0.1259        16       416: 100% 12/12 [00:02<00:00,  5.71it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.49it/s]
                 all          53          54           0           0      0.0119     0.00201

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     49/99     1.42G    0.0643   0.06599         0    0.1303        19       416: 100% 12/12 [00:02<00:00,  5.80it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.39it/s]
                 all          53          54     0.00301      0.0556      0.0116     0.00202

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     50/99     1.42G   0.06459   0.06228         0    0.1269        20       416: 100% 12/12 [00:02<00:00,  5.76it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.48it/s]
                 all          53          54     0.00423      0.0926      0.0117     0.00203

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     51/99     1.42G   0.06315   0.06119         0    0.1243        21       416: 100% 12/12 [00:02<00:00,  5.66it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.57it/s]
                 all          53          54     0.00354      0.0741      0.0138     0.00256

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     52/99     1.42G   0.06466   0.06232         0     0.127        18       416: 100% 12/12 [00:02<00:00,  5.62it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.35it/s]
                 all          53          54      0.0144      0.0741      0.0279     0.00728

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     53/99     1.42G    0.0651   0.06484         0    0.1299        24       416: 100% 12/12 [00:02<00:00,  5.71it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.50it/s]
                 all          53          54      0.0217      0.0926      0.0272     0.00502

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     54/99     1.42G   0.06062   0.06164         0    0.1223        14       416: 100% 12/12 [00:02<00:00,  5.87it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.55it/s]
                 all          53          54       0.042       0.037      0.0247     0.00616

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     55/99     1.42G   0.06653   0.06422         0    0.1308        21       416: 100% 12/12 [00:02<00:00,  5.66it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.33it/s]
                 all          53          54           0           0      0.0267     0.00555

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     56/99     1.42G   0.06236   0.06298         0    0.1253        21       416: 100% 12/12 [00:02<00:00,  5.67it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.53it/s]
                 all          53          54           0           0      0.0442     0.00971

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     57/99     1.42G   0.06225   0.06343         0    0.1257        21       416: 100% 12/12 [00:02<00:00,  5.70it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.49it/s]
                 all          53          54           0           0       0.065       0.017

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     58/99     1.42G   0.06119   0.06261         0    0.1238        22       416: 100% 12/12 [00:02<00:00,  5.78it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.35it/s]
                 all          53          54           0           0      0.0417      0.0104

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     59/99     1.42G   0.06067   0.05844         0    0.1191        24       416: 100% 12/12 [00:02<00:00,  5.74it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.54it/s]
                 all          53          54           0           0       0.031     0.00699

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     60/99     1.42G   0.06369   0.06742         0    0.1311        21       416: 100% 12/12 [00:02<00:00,  5.71it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.61it/s]
                 all          53          54           0           0      0.0306     0.00505

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     61/99     1.42G   0.06407   0.06509         0    0.1292        25       416: 100% 12/12 [00:02<00:00,  5.70it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.54it/s]
                 all          53          54      0.0706       0.037      0.0296     0.00531

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     62/99     1.42G   0.06329     0.062         0    0.1253        21       416: 100% 12/12 [00:02<00:00,  5.81it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.44it/s]
                 all          53          54           0           0      0.0272     0.00494

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     63/99     1.42G   0.06046   0.06243         0    0.1229        25       416: 100% 12/12 [00:02<00:00,  5.78it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.53it/s]
                 all          53          54           0           0      0.0333     0.00617

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     64/99     1.42G   0.05635   0.05767         0     0.114        18       416: 100% 12/12 [00:02<00:00,  5.76it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.52it/s]
                 all          53          54           0           0        0.04     0.00581

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     65/99     1.42G   0.06252   0.06835         0    0.1309        25       416: 100% 12/12 [00:02<00:00,  5.73it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.35it/s]
                 all          53          54      0.0463       0.037      0.0511     0.00822

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     66/99     1.42G    0.0612    0.0665         0    0.1277        21       416: 100% 12/12 [00:02<00:00,  5.70it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.53it/s]
                 all          53          54      0.0112       0.037      0.0404     0.00708

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     67/99     1.42G   0.05886   0.06563         0    0.1245        19       416: 100% 12/12 [00:02<00:00,  5.71it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.52it/s]
                 all          53          54      0.0145       0.037      0.0382     0.00652

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     68/99     1.42G    0.0626   0.06934         0    0.1319        23       416: 100% 12/12 [00:02<00:00,  5.70it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.34it/s]
                 all          53          54       0.021      0.0185      0.0351     0.00548

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     69/99     1.42G   0.05777   0.06507         0    0.1228        20       416: 100% 12/12 [00:02<00:00,  5.89it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.50it/s]
                 all          53          54           0           0      0.0328     0.00502

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     70/99     1.42G   0.06121   0.06384         0    0.1251        23       416: 100% 12/12 [00:02<00:00,  5.92it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.55it/s]
                 all          53          54           0           0      0.0178     0.00295

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     71/99     1.42G   0.06121   0.06745         0    0.1287        28       416: 100% 12/12 [00:02<00:00,  5.85it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.37it/s]
                 all          53          54      0.0245      0.0185      0.0305     0.00519

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     72/99     1.42G    0.0587   0.06349         0    0.1222        17       416: 100% 12/12 [00:02<00:00,  5.84it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.51it/s]
                 all          53          54      0.0175      0.0185      0.0328     0.00534

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     73/99     1.42G   0.06023   0.06595         0    0.1262        19       416: 100% 12/12 [00:02<00:00,  5.80it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.58it/s]
                 all          53          54           0           0      0.0246     0.00366

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     74/99     1.42G   0.05783     0.065         0    0.1228        18       416: 100% 12/12 [00:02<00:00,  5.92it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.53it/s]
                 all          53          54           0           0      0.0415     0.00572

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     75/99     1.42G   0.05807   0.05998         0    0.1181        24       416: 100% 12/12 [00:02<00:00,  5.91it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.37it/s]
                 all          53          54           0           0      0.0468     0.00728

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     76/99     1.42G   0.05977    0.0651         0    0.1249        21       416: 100% 12/12 [00:02<00:00,  5.76it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.58it/s]
                 all          53          54           0           0      0.0305     0.00562

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     77/99     1.42G   0.05414   0.06481         0     0.119        29       416: 100% 12/12 [00:02<00:00,  5.84it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.59it/s]
                 all          53          54           0           0      0.0407     0.00601

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     78/99     1.42G      0.06   0.06295         0     0.123        26       416: 100% 12/12 [00:02<00:00,  5.73it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.40it/s]
                 all          53          54           0           0      0.0302     0.00507

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     79/99     1.42G   0.05838   0.05901         0    0.1174        21       416: 100% 12/12 [00:02<00:00,  5.84it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.56it/s]
                 all          53          54           0           0      0.0306      0.0054

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     80/99     1.42G   0.06153   0.06338         0    0.1249        21       416: 100% 12/12 [00:02<00:00,  5.88it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.48it/s]
                 all          53          54           0           0      0.0385     0.00649

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     81/99     1.42G   0.05934   0.06757         0    0.1269        23       416: 100% 12/12 [00:02<00:00,  5.88it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.43it/s]
                 all          53          54       0.024      0.0185      0.0518     0.00743

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     82/99     1.42G   0.05941   0.06107         0    0.1205        23       416: 100% 12/12 [00:02<00:00,  5.91it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.59it/s]
                 all          53          54      0.0348      0.0321      0.0567     0.00864

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     83/99     1.42G    0.0574   0.06215         0    0.1195        23       416: 100% 12/12 [00:02<00:00,  5.80it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.52it/s]
                 all          53          54      0.0429       0.037       0.053     0.00872

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     84/99     1.42G   0.05506   0.06326         0    0.1183        19       416: 100% 12/12 [00:02<00:00,  5.82it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.57it/s]
                 all          53          54      0.0418       0.037      0.0422     0.00646

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     85/99     1.42G   0.05627    0.0635         0    0.1198        15       416: 100% 12/12 [00:02<00:00,  5.89it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.39it/s]
                 all          53          54      0.0239      0.0185      0.0515     0.00852

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     86/99     1.42G    0.0606   0.06497         0    0.1256        18       416: 100% 12/12 [00:02<00:00,  5.84it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.59it/s]
                 all          53          54      0.0239      0.0185      0.0413     0.00714

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     87/99     1.42G    0.0577   0.06326         0     0.121        12       416: 100% 12/12 [00:02<00:00,  5.85it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.61it/s]
                 all          53          54      0.0477      0.0556      0.0514     0.00899

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     88/99     1.42G   0.05719   0.06302         0    0.1202        27       416: 100% 12/12 [00:02<00:00,  5.81it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.33it/s]
                 all          53          54      0.0308      0.0556      0.0406     0.00733

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     89/99     1.42G   0.05494   0.06097         0    0.1159        21       416: 100% 12/12 [00:02<00:00,  5.84it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.54it/s]
                 all          53          54      0.0397      0.0556      0.0386     0.00725

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     90/99     1.42G   0.05738   0.06357         0    0.1209        17       416: 100% 12/12 [00:02<00:00,  5.85it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.54it/s]
                 all          53          54       0.082       0.111      0.0503     0.00709

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     91/99     1.42G   0.05486    0.0618         0    0.1167        24       416: 100% 12/12 [00:02<00:00,  5.79it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.39it/s]
                 all          53          54      0.0813      0.0926      0.0602     0.00927

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     92/99     1.42G   0.05755   0.06544         0     0.123        19       416: 100% 12/12 [00:02<00:00,  5.77it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.57it/s]
                 all          53          54      0.0553      0.0556      0.0534      0.0098

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     93/99     1.42G   0.05647   0.05828         0    0.1148        19       416: 100% 12/12 [00:02<00:00,  5.67it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.61it/s]
                 all          53          54      0.0774       0.111      0.0529      0.0107

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     94/99     1.42G   0.05591   0.06317         0    0.1191        21       416: 100% 12/12 [00:02<00:00,  5.83it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.53it/s]
                 all          53          54       0.108       0.211      0.0656      0.0123

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     95/99     1.42G   0.05442    0.0667         0    0.1211        16       416: 100% 12/12 [00:02<00:00,  5.91it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.39it/s]
                 all          53          54       0.107       0.181      0.0603     0.00995

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     96/99     1.42G   0.05688   0.06263         0    0.1195        20       416: 100% 12/12 [00:02<00:00,  5.86it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.50it/s]
                 all          53          54      0.0481      0.0926       0.049     0.00823

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     97/99     1.42G   0.05705   0.06476         0    0.1218        25       416: 100% 12/12 [00:02<00:00,  5.95it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.54it/s]
                 all          53          54      0.0768       0.167      0.0571     0.00867

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     98/99     1.42G    0.0585   0.06253         0     0.121        31       416: 100% 12/12 [00:02<00:00,  5.88it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:01<00:00,  2.39it/s]
                 all          53          54      0.0685       0.148      0.0532     0.00806

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
     99/99     1.42G   0.05563    0.0677         0    0.1233        24       416: 100% 12/12 [00:02<00:00,  5.91it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 4/4 [00:03<00:00,  1.30it/s]
                 all          53          54       0.088       0.259      0.0588     0.00847
Optimizer stripped from runs/exp0_yolov5s_results/weights/last.pt, 14.7MB
Optimizer stripped from runs/exp0_yolov5s_results/weights/best.pt, 14.7MB
100 epochs completed in 0.114 hours.

CPU times: user 2.36 s, sys: 445 ms, total: 2.8 s
Wall time: 7min 15s
Unfortunately this also did NOT produce any successful detections. This might suggest that there is nothing fundamentally wrong with our setup but we simply need either better training data or more training data.
Daniel Wee
Site Admin
Posts: 2245
Joined: Wed 25 Feb 25 2009 8:00 pm

Re: YoloV5 recognition

Post by Daniel Wee » Sat 05 Sep 05 2020 4:48 pm

We will test the training data against a YOLOv4 implementation on Roboflow:-

https://colab.research.google.com/drive ... lhqP2S57Ub

I made a copy of the successful run with changes made to get it running:-

https://colab.research.google.com/drive ... JS7khL3HLI

This WORKED. Inference was possible so there is nothing fundamentally wrong with the training data set. It might be the requirements that were not met in the pytorch installation.
Daniel Wee
Site Admin
Posts: 2245
Joined: Wed 25 Feb 25 2009 8:00 pm

Re: YOLO v5 recognition

Post by Daniel Wee » Sat 05 Sep 05 2020 5:58 pm

We finally got a breakthrough!!!!

The Colab version has been trained and is producing some inference results!

https://colab.research.google.com/drive ... xi2ncxWffW

This was done with the training set located in:-

http://www.tsebi.com/reviews/Lizards.v1 ... ytorch.zip

We will now increase this dataset to correspond with what we have been using locally to see if that creates any problems. We will also train this to 300 epochs to try and get better detection results. The new training set will be:-

http://www.tsebi.com/reviews/Lizards.v2 ... ytorch.zip

I am noticing that the file structure produced by Roboflow is different from the one provided by the examples. We will need to investigate this difference to see if that results in any improvement with our local training setup.
Daniel Wee
Site Admin
Posts: 2245
Joined: Wed 25 Feb 25 2009 8:00 pm

Re: YOLO v5 recognition

Post by Daniel Wee » Sat 05 Sep 05 2020 8:41 pm

The next step would be to generate a new set of test images from lizard3.mp4 which has many small targets.

We tried with this website:-

https://ezgif.com/video-to-jpg/ezgif-2-a72b70ae269c.mp4

This gave us 30-seconds of 5-fps frames but the quality were quite low.

I will try again using VLC player's scene filter per these instructions:-

https://www.raymond.cc/blog/extract-vid ... ia-player/
Daniel Wee
Site Admin
Posts: 2245
Joined: Wed 25 Feb 25 2009 8:00 pm

Re: YOLO v5 recognition

Post by Daniel Wee » Sun 06 Sep 06 2020 12:01 am

Using the Colab setup, I retrained the model further using the previous best.pt as basis with much smaller targets. The resultant model (best2.pt) was good for recognizing smaller lizards but did not identify larger images as best.pt managed to do. I think we need to mix in all the images and train them as one large group rather than sequentially. We will need to find out how to further train a model.

Code: Select all

python detect.py --weights weights/best2.pt --source inference/videos/lizard3.mp4 --view-img --output inference/output
Daniel Wee
Site Admin
Posts: 2245
Joined: Wed 25 Feb 25 2009 8:00 pm

Re: YOLO v5 recognition

Post by Daniel Wee » Sun 06 Sep 06 2020 1:08 pm

I am using the training file structure that Roboflow provides, and this is now found in yolov5/training2 and the training command is:-

Code: Select all

python train.py --img 640 --batch 16 --epochs 300 --data training2/data.yaml --cfg training2/yolov5l.yaml --weights ''
Hopefully this yields usable trained weights on the local machine.

Result: FAIL

The local test still did not yield files that produced any inference result in spite of the new structure used along with the new data.yaml in yolov5/training2. This is increasingly looking like it might be a pytorch or CUDA issue. We will need to re-install some of that to be sure.

Meanwhile the expanded training set tested and worked well. I am increasing the training data and testing to 600 epochs this time. We were assigned a Tesla P4 with only about 7.6GB RAM so we had to reduce the batch size to 10. This might improve generalization by increasing noise though.
Daniel Wee
Site Admin
Posts: 2245
Joined: Wed 25 Feb 25 2009 8:00 pm

Re: YOLO v5 recognition

Post by Daniel Wee » Sun 06 Sep 06 2020 10:49 pm

I am re-installing CUDA 10.1 and cuDNN 7.6.5.32 for CUDA 10.1 following the instructions here:-

https://towardsdatascience.com/installi ... 693e46e781

We may have to try this on an Ubuntu installation - it could be a Windows problem with YOLO v5.

Result: SUCCESS!

Running the training on the local Ubuntu 20.04 setup worked! It looks like the problem has to do with the Windows installation of pytorch?
Post Reply