yolov5_birdproject/tutorial.ipynb

This notebook contains software developed by Ultralytics LLC, and is freely available for redistribution under the GPL-3.0 license. For more information please visit https://github.com/ultralytics/yolov5 and https://www.ultralytics.com.

#Initial Setup

Clone repo, install dependencies and RESTART RUNTIME.

In [0]:

!git clone https://github.com/ultralytics/yolov5  # clone repo
!pip install -U -r yolov5/requirements.txt  # install dependencies and RESTART RUNTIME

After restarting runtime, %cd back into ./yolov5 folder, import torch and check CUDA device.

In [64]:

%cd yolov5

/content/yolov5

In [66]:

import torch
from IPython.display import Image  # for displaying images
from utils.google_utils import gdrive_download  # for downloading models/datasets

print('torch %s %s' % (torch.__version__, torch.cuda.get_device_properties(0) if torch.cuda.is_available() else 'CPU'))

torch 1.5.0+cu101 _CudaDeviceProperties(name='Tesla P100-PCIE-16GB', major=6, minor=0, total_memory=16280MB, multi_processor_count=56)

In [0]:

# Re-clone (only if necessary, not normally used)
%cd ..
!rm -rf yolov5 && git clone https://github.com/ultralytics/yolov5
%cd yolov5

#1. Inference

Run inference with a pretrained checkpoint on contents of inference/images folder. Models are downloaded automatically from our Google Drive folder if available.

In [0]:

# Run inference
!python detect.py --weights yolov5s.pt --img 640 --conf 0.4 --source ./inference/images/
Image(filename='inference/output/zidane.jpg', width=600)

Out [0]:

Namespace(agnostic_nms=False, augment=False, classes=None, conf_thres=0.4, device='', fourcc='mp4v', half=False, img_size=640, iou_thres=0.5, output='inference/output', save_txt=False, source='./inference/images/', view_img=False, weights='yolov5s.pt')
Using CUDA device0 _CudaDeviceProperties(name='Tesla P100-PCIE-16GB', total_memory=16280MB)

image 1/2 inference/images/bus.jpg: 640x512 3 persons, 1 buss, Done. (0.009s)
image 2/2 inference/images/zidane.jpg: 384x640 2 persons, 2 ties, Done. (0.009s)
Results saved to /content/yolov5/inference/output
Done. (0.118s)

#2. Test Test a model on COCO val or test-dev dataset to determine trained accuracy. Models are downloaded automatically from our Google Drive folder if available. To show results by class use the --verbose flag. Note that pycocotools metrics may be 1-2% better than the equivalent repo metrics, as is visible below, due to slight differences in mAP computation.

###2.1 val2017 Download COCO val 2017 dataset, 1GB, 5000 images, and test model accuracy.

In [0]:

# Download COCO val2017
gdrive_download('1Y6Kou6kEB0ZEMCCpJSKStCor4KAReE43','coco2017val.zip')  # val2017 dataset
!mv ./coco ../  # move folder alongside /yolov5

Downloading https://drive.google.com/uc?export=download&id=1Y6Kou6kEB0ZEMCCpJSKStCor4KAReE43 as coco2017val.zip... unzipping... Done (25.8s)

In [56]:

# Run yolov5s on COCO val2017
!python test.py --weights yolov5s.pt --data ./data/coco.yaml

Namespace(augment=False, batch_size=16, conf_thres=0.001, data='././data/coco.yaml', device='', img_size=640, iou_thres=0.65, save_json=True, single_cls=False, task='val', verbose=False, weights='yolov5s.pt')
Using CUDA device0 _CudaDeviceProperties(name='Tesla P100-PCIE-16GB', total_memory=16280MB)

Downloading https://drive.google.com/uc?export=download&id=1R5T6rIyy3lLwgFXNms8whc-387H0tMQO as yolov5s.pt... Done (1.9s)
Model Summary: 99 layers, 6.99302e+06 parameters, 6.99302e+06 gradients
Caching labels ../coco/labels/val2017.npy (4952 found, 0 missing, 48 empty, 0 duplicate, for 5000 images): 100% 5000/5000 [00:00<00:00, 23176.56it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 313/313 [01:41<00:00,  3.07it/s]
                 all       5e+03    3.63e+04       0.317       0.621        0.52       0.314
Speed: 3.1/2.8/5.8 ms inference/NMS/total per 640x640 image at batch-size 16

COCO mAP with pycocotools... saving detections_val2017_yolov5s_results.json...
loading annotations into memory...
Done (t=0.40s)
creating index...
index created!
Loading and preparing results...
DONE (t=13.01s)
creating index...
index created!
Running per image evaluation...
Evaluate annotation type *bbox*
DONE (t=130.84s).
Accumulating evaluation results...
DONE (t=24.98s).
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.330
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.528
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.350
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.185
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.374
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.422
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.282
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.482
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.558
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.397
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.618
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.691

In [0]:

# Run multiple models on COCO val2017
%%shell
for x in yolov5s yolov5m yolov5l yolov5x yolov3-spp
do 
  python test.py --weights $x.pt --data ./data/coco.yaml
done

###2.2 test-dev2017 Download COCO test2017 dataset, 7GB, 40,000 images, to test model accuracy on test-dev set, 20,000 images. Results are saved to a *.json file which can be submitted to the evaluation server at https://competitions.codalab.org/competitions/20794.

In [0]:

# Download COCO test-dev2017
gdrive_download('1cXZR_ckHki6nddOmcysCuuJFM--T-Q6L','coco2017labels.zip')  # annotations
!f="test2017.zip" && curl http://images.cocodataset.org/zips/$f -o $f && unzip -q $f && rm $f  # 7GB,  41k images
!mv ./test2017 ./coco/images && mv ./coco ../  # move images into /coco and move /coco alongside /yolov5

Downloading https://drive.google.com/uc?export=download&id=1cXZR_ckHki6nddOmcysCuuJFM--T-Q6L as coco2017labels.zip... unzipping... Done (9.3s)
  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed
100 6339M  100 6339M    0     0  16.2M      0  0:06:30  0:06:30 --:--:-- 16.3M

In [0]:

# Run yolov5s on COCO test-dev2017
!python test.py --weights yolov5s.pt --data ./data/coco.yaml --task test

3. Train

Download the 128-image tutorial training dataset ./data/coco128.yaml, start tensorboard and train a yolov5s.yaml model for 5 epochs. Note that actual training is typically much longer, around 300-1000 epochs, depending on your dataset.

In [57]:

# Download tutorial dataset coco128.yaml
gdrive_download('1n_oKgR81BJtqk75b00eAjdv03qVCQn2f','coco128.zip')  # tutorial dataset
!mv ./coco128 ../  # move folder alongside /yolov5

Downloading https://drive.google.com/uc?export=download&id=1n_oKgR81BJtqk75b00eAjdv03qVCQn2f as coco128.zip... unzipping... Done (4.8s)

In [0]:

# Start tensorboard
%load_ext tensorboard
%tensorboard --logdir runs

Train a yolov5s model on the coco128 dataset by specifying model configuration file --cfg ./models/yolo5s.yaml, and a dataset configuration file --data ./data/coco128.yaml. Start training from pretrained --weights yolov5s.pt, or from scratch (randomly initialized weights) using --weights ''

Models are downloaded automatically from our Google Drive folder if available.

In [62]:

# Train yolov5s on coco128 for 5 epochs
!python train.py --img 640 --batch 16 --epochs 5 --data ./data/coco128.yaml --cfg ./models/yolov5s.yaml --weights '' --name yolov5s_coco128 --nosave --cache

Apex recommended for faster mixed precision training: https://github.com/NVIDIA/apex
{'lr0': 0.01, 'momentum': 0.937, 'weight_decay': 0.0005, 'giou': 0.05, 'cls': 0.58, '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': 0.014, 'hsv_s': 0.68, 'hsv_v': 0.36, 'degrees': 0.0, 'translate': 0.0, 'scale': 0.5, 'shear': 0.0}
Namespace(adam=False, batch_size=16, bucket='', cache_images=True, cfg='././models/yolov5s.yaml', data='././data/coco128.yaml', device='', epochs=5, evolve=False, img_size=[640], name='yolov5s_coco128', nosave=True, notest=False, rect=False, resume=False, single_cls=False, weights='')
Using CUDA device0 _CudaDeviceProperties(name='Tesla P100-PCIE-16GB', total_memory=16280MB)

2020-05-30 00:01:16.653829: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudart.so.10.1
Start Tensorboard with "tensorboard --logdir=runs", view at http://localhost:6006/

              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     20672  models.common.Bottleneck                [64, 64]                      
  3             -1  1     73984  models.common.Conv                      [64, 128, 3, 2]               
  4             -1  3    246912  models.common.Bottleneck                [128, 128]                    
  5             -1  1    295424  models.common.Conv                      [128, 256, 3, 2]              
  6             -1  3    985344  models.common.Bottleneck                [256, 256]                    
  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   1312256  models.common.Bottleneck                [512, 512]                    
 10             -1  1   1312256  models.common.Bottleneck                [512, 512, False]             
 11             -1  1    130815  torch.nn.modules.conv.Conv2d            [512, 255, 1, 1, 0]           
 12             -2  1         0  torch.nn.modules.upsampling.Upsample    [None, 2, 'nearest']          
 13        [-1, 6]  1         0  models.common.Concat                    [1]                           
 14             -1  1    197120  models.common.Conv                      [768, 256, 1, 1]              
 15             -1  1    328448  models.common.Bottleneck                [256, 256, False]             
 16             -1  1     65535  torch.nn.modules.conv.Conv2d            [256, 255, 1, 1, 0]           
 17             -2  1         0  torch.nn.modules.upsampling.Upsample    [None, 2, 'nearest']          
 18        [-1, 4]  1         0  models.common.Concat                    [1]                           
 19             -1  1     49408  models.common.Conv                      [384, 128, 1, 1]              
 20             -1  1     82304  models.common.Bottleneck                [128, 128, False]             
 21             -1  1     32895  torch.nn.modules.conv.Conv2d            [128, 255, 1, 1, 0]           
 22   [-1, 16, 11]  1         0  models.yolo.Detect                      [80, [[10, 13, 16, 30, 33, 23], [30, 61, 62, 45, 59, 119], [116, 90, 156, 198, 373, 326]]]
Model Summary: 99 layers, 6.99302e+06 parameters, 6.99302e+06 gradients

Optimizer groups: 34 .bias, 34 conv.weight, 31 other
Caching labels ../coco128/labels/train2017 (126 found, 0 missing, 2 empty, 0 duplicate, for 128 images): 100% 128/128 [00:00<00:00, 6595.06it/s]
Caching images (0.1GB): 100% 128/128 [00:00<00:00, 154.24it/s]
Caching labels ../coco128/labels/train2017 (126 found, 0 missing, 2 empty, 0 duplicate, for 128 images): 100% 128/128 [00:00<00:00, 9849.04it/s]
Caching images (0.1GB): 100% 128/128 [00:00<00:00, 151.99it/s]
Image sizes 640 train, 640 test
Using 2 dataloader workers
Starting training for 5 epochs...

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
       0/4     9.65G    0.1222   0.08038    0.1221    0.3248       223       640: 100% 8/8 [00:06<00:00,  1.14it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 8/8 [00:03<00:00,  2.31it/s]
                 all         128         929           0           0           0           0

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
       1/4     8.86G    0.1163   0.08987    0.1211    0.3273       348       640: 100% 8/8 [00:02<00:00,  3.64it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 8/8 [00:00<00:00,  8.80it/s]
                 all         128         929           0           0           0           0

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
       2/4     8.86G    0.1145   0.08071    0.1214    0.3166       187       640: 100% 8/8 [00:02<00:00,  3.62it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 8/8 [00:00<00:00,  8.62it/s]
                 all         128         929           0           0           0           0

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
       3/4     8.86G     0.113   0.07988    0.1199    0.3128       236       640: 100% 8/8 [00:02<00:00,  3.53it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 8/8 [00:00<00:00,  8.79it/s]
                 all         128         929           0           0           0           0

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
       4/4     8.86G    0.1109   0.08885    0.1198    0.3196       257       640: 100% 8/8 [00:02<00:00,  3.62it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 8/8 [00:00<00:00,  8.56it/s]
                 all         128         929           0           0           0           0
Optimizer stripped from weights/last_yolov5s_coco128.pt
5 epochs completed in 0.007 hours.

#4. Visualize

After training starts, view train*.jpg images to see training images, labels and augmentation effects. Note a mosaic dataloader is used for training (shown below), a new dataloading concept developed by Ultralytics and first featured in YOLOv4.

In [0]:

Image(filename='./train_batch1.jpg', width=900)  # view augmented train images

Out [0]:

View test_batch0_gt.jpg to see test batch 0 ground truth labels.

In [0]:

Image(filename='./test_batch0_gt.jpg', width=900)  # view test image labels

Out [0]:

View test_batch0_pred.jpg to see test batch 0 predictions.

In [0]:

Image(filename='./test_batch0_pred.jpg', width=900)  # view test image predictions

Out [0]:

Training losses and performance metrics are saved to Tensorboard and also to a results.txt logfile. results.txt is plotted as results.png after training completes. Partially completed results.txt files can be plotted with from utils.utils import plot_results; plot_results(). Here we show yolov5s trained on coco128 to 100 epochs, starting from scratch (blue), and starting from pretrained yolov5s.pt weights (orange).

In [0]:

from utils.utils import plot_results; plot_results()  # plot results.txt as results.png
Image(filename='./results.png', width=1000)  # view results.png

Out [0]: