57 KiB
Vendored
57 KiB
Vendored
This is the official YOLOv5 🚀 notebook by Ultralytics, and is freely available for redistribution under the GPL-3.0 license. For more information please visit https://github.com/ultralytics/yolov5 and https://ultralytics.com. Thank you!
In [ ]:
!git clone https://github.com/ultralytics/yolov5 # clone
%cd yolov5
%pip install -qr requirements.txt # install
import torch
import utils
display = utils.notebook_init() # checksYOLOv5 🚀 v6.2-15-g61adf01 Python-3.7.13 torch-1.12.1+cu113 CUDA:0 (Tesla V100-SXM2-16GB, 16160MiB)
Setup complete ✅ (8 CPUs, 51.0 GB RAM, 37.4/166.8 GB disk)
In [ ]:
!python detect.py --weights yolov5s.pt --img 640 --conf 0.25 --source data/images
# display.Image(filename='runs/detect/exp/zidane.jpg', width=600)[34m[1mdetect: [0mweights=['yolov5s.pt'], source=data/images, data=data/coco128.yaml, imgsz=[640, 640], conf_thres=0.25, iou_thres=0.45, max_det=1000, device=, view_img=False, save_txt=False, save_conf=False, save_crop=False, nosave=False, classes=None, agnostic_nms=False, augment=False, visualize=False, update=False, project=runs/detect, name=exp, exist_ok=False, line_thickness=3, hide_labels=False, hide_conf=False, half=False, dnn=False YOLOv5 🚀 v6.2-15-g61adf01 Python-3.7.13 torch-1.12.1+cu113 CUDA:0 (Tesla V100-SXM2-16GB, 16160MiB) Downloading https://github.com/ultralytics/yolov5/releases/download/v6.2/yolov5s.pt to yolov5s.pt... 100% 14.1M/14.1M [00:00<00:00, 39.3MB/s] Fusing layers... YOLOv5s summary: 213 layers, 7225885 parameters, 0 gradients image 1/2 /content/yolov5/data/images/bus.jpg: 640x480 4 persons, 1 bus, 14.9ms image 2/2 /content/yolov5/data/images/zidane.jpg: 384x640 2 persons, 2 ties, 22.0ms Speed: 0.6ms pre-process, 18.4ms inference, 24.1ms NMS per image at shape (1, 3, 640, 640) Results saved to [1mruns/detect/exp[0m
In [ ]:
# Download COCO val
torch.hub.download_url_to_file('https://ultralytics.com/assets/coco2017val.zip', 'tmp.zip') # download COCO val (1GB - 5000 images)
!unzip -q tmp.zip -d ../datasets && rm tmp.zip # unzip0%| | 0.00/780M [00:00<?, ?B/s]
In [ ]:
# Validate YOLOv5x on COCO val
!python val.py --weights yolov5x.pt --data coco.yaml --img 640 --iou 0.65 --half[34m[1mval: [0mdata=/content/yolov5/data/coco.yaml, weights=['yolov5x.pt'], batch_size=32, imgsz=640, conf_thres=0.001, iou_thres=0.65, task=val, device=, workers=8, single_cls=False, augment=False, verbose=False, save_txt=False, save_hybrid=False, save_conf=False, save_json=True, project=runs/val, name=exp, exist_ok=False, half=True, dnn=False
YOLOv5 🚀 v6.2-15-g61adf01 Python-3.7.13 torch-1.12.1+cu113 CUDA:0 (Tesla V100-SXM2-16GB, 16160MiB)
Downloading https://github.com/ultralytics/yolov5/releases/download/v6.2/yolov5x.pt to yolov5x.pt...
100% 166M/166M [00:06<00:00, 28.1MB/s]
Fusing layers...
YOLOv5x summary: 444 layers, 86705005 parameters, 0 gradients
Downloading https://ultralytics.com/assets/Arial.ttf to /root/.config/Ultralytics/Arial.ttf...
100% 755k/755k [00:00<00:00, 47.3MB/s]
[34m[1mval: [0mScanning '/content/datasets/coco/val2017' images and labels...4952 found, 48 missing, 0 empty, 0 corrupt: 100% 5000/5000 [00:00<00:00, 10756.32it/s]
[34m[1mval: [0mNew cache created: /content/datasets/coco/val2017.cache
Class Images Labels P R mAP@.5 mAP@.5:.95: 100% 157/157 [01:07<00:00, 2.33it/s]
all 5000 36335 0.743 0.625 0.683 0.504
Speed: 0.1ms pre-process, 4.6ms inference, 1.2ms NMS per image at shape (32, 3, 640, 640)
Evaluating pycocotools mAP... saving runs/val/exp/yolov5x_predictions.json...
loading annotations into memory...
Done (t=0.41s)
creating index...
index created!
Loading and preparing results...
DONE (t=5.64s)
creating index...
index created!
Running per image evaluation...
Evaluate annotation type *bbox*
DONE (t=76.80s).
Accumulating evaluation results...
DONE (t=14.61s).
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.506
Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.688
Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.549
Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.340
Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.558
Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.651
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.382
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.631
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.684
Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.528
Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.737
Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.833
Results saved to [1mruns/val/exp[0m
3. Train
Train a YOLOv5s model on the COCO128 dataset with --data coco128.yaml, starting from pretrained --weights yolov5s.pt, or from randomly initialized --weights '' --cfg yolov5s.yaml.
- Pretrained Models are downloaded automatically from the latest YOLOv5 release
- Datasets available for autodownload include: COCO, COCO128, VOC, Argoverse, VisDrone, GlobalWheat, xView, Objects365, SKU-110K.
- Training Results are saved to
runs/train/with incrementing run directories, i.e.runs/train/exp2,runs/train/exp3etc.
A Mosaic Dataloader is used for training which combines 4 images into 1 mosaic.
Train on Custom Data with Roboflow 🌟 NEW
Roboflow enables you to easily organize, label, and prepare a high quality dataset with your own custom data. Roboflow also makes it easy to establish an active learning pipeline, collaborate with your team on dataset improvement, and integrate directly into your model building workflow with the roboflow pip package.
- Custom Training Example: https://blog.roboflow.com/how-to-train-yolov5-on-a-custom-dataset/
- Custom Training Notebook:
In [ ]:
# Tensorboard (optional)
%load_ext tensorboard
%tensorboard --logdir runs/trainIn [ ]:
# ClearML (optional)
%pip install -q clearml
!clearml-initIn [ ]:
# Weights & Biases (optional)
%pip install -q wandb
import wandb
wandb.login()In [ ]:
# Train YOLOv5s on COCO128 for 3 epochs
!python train.py --img 640 --batch 16 --epochs 3 --data coco128.yaml --weights yolov5s.pt --cache[34m[1mtrain: [0mweights=yolov5s.pt, cfg=, data=coco128.yaml, hyp=data/hyps/hyp.scratch-low.yaml, epochs=3, batch_size=16, imgsz=640, rect=False, resume=False, nosave=False, noval=False, noautoanchor=False, noplots=False, evolve=None, bucket=, cache=ram, image_weights=False, device=, multi_scale=False, single_cls=False, optimizer=SGD, sync_bn=False, workers=8, project=runs/train, name=exp, exist_ok=False, quad=False, cos_lr=False, label_smoothing=0.0, patience=100, freeze=[0], save_period=-1, seed=0, local_rank=-1, entity=None, upload_dataset=False, bbox_interval=-1, artifact_alias=latest
[34m[1mgithub: [0mup to date with https://github.com/ultralytics/yolov5 ✅
YOLOv5 🚀 v6.2-15-g61adf01 Python-3.7.13 torch-1.12.1+cu113 CUDA:0 (Tesla V100-SXM2-16GB, 16160MiB)
[34m[1mhyperparameters: [0mlr0=0.01, lrf=0.01, momentum=0.937, weight_decay=0.0005, warmup_epochs=3.0, warmup_momentum=0.8, warmup_bias_lr=0.1, box=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, mosaic=1.0, mixup=0.0, copy_paste=0.0
[34m[1mWeights & Biases: [0mrun 'pip install wandb' to automatically track and visualize YOLOv5 🚀 runs in Weights & Biases
[34m[1mClearML: [0mrun 'pip install clearml' to automatically track, visualize and remotely train YOLOv5 🚀 in ClearML
[34m[1mTensorBoard: [0mStart with 'tensorboard --logdir runs/train', view at http://localhost:6006/
Dataset not found ⚠️, missing paths ['/content/datasets/coco128/images/train2017']
Downloading https://ultralytics.com/assets/coco128.zip to coco128.zip...
100% 6.66M/6.66M [00:00<00:00, 75.3MB/s]
Dataset download success ✅ (0.7s), saved to [1m/content/datasets[0m
from n params module arguments
0 -1 1 3520 models.common.Conv [3, 32, 6, 2, 2]
1 -1 1 18560 models.common.Conv [32, 64, 3, 2]
2 -1 1 18816 models.common.C3 [64, 64, 1]
3 -1 1 73984 models.common.Conv [64, 128, 3, 2]
4 -1 2 115712 models.common.C3 [128, 128, 2]
5 -1 1 295424 models.common.Conv [128, 256, 3, 2]
6 -1 3 625152 models.common.C3 [256, 256, 3]
7 -1 1 1180672 models.common.Conv [256, 512, 3, 2]
8 -1 1 1182720 models.common.C3 [512, 512, 1]
9 -1 1 656896 models.common.SPPF [512, 512, 5]
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 361984 models.common.C3 [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 90880 models.common.C3 [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 296448 models.common.C3 [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 1182720 models.common.C3 [512, 512, 1, False]
24 [17, 20, 23] 1 229245 models.yolo.Detect [80, [[10, 13, 16, 30, 33, 23], [30, 61, 62, 45, 59, 119], [116, 90, 156, 198, 373, 326]], [128, 256, 512]]
Model summary: 270 layers, 7235389 parameters, 7235389 gradients, 16.6 GFLOPs
Transferred 349/349 items from yolov5s.pt
[34m[1mAMP: [0mchecks passed ✅
[34m[1moptimizer:[0m SGD(lr=0.01) with parameter groups 57 weight(decay=0.0), 60 weight(decay=0.0005), 60 bias
[34m[1malbumentations: [0mBlur(p=0.01, blur_limit=(3, 7)), MedianBlur(p=0.01, blur_limit=(3, 7)), ToGray(p=0.01), CLAHE(p=0.01, clip_limit=(1, 4.0), tile_grid_size=(8, 8))
[34m[1mtrain: [0mScanning '/content/datasets/coco128/labels/train2017' images and labels...128 found, 0 missing, 2 empty, 0 corrupt: 100% 128/128 [00:00<00:00, 7246.20it/s]
[34m[1mtrain: [0mNew cache created: /content/datasets/coco128/labels/train2017.cache
[34m[1mtrain: [0mCaching images (0.1GB ram): 100% 128/128 [00:00<00:00, 986.21it/s]
[34m[1mval: [0mScanning '/content/datasets/coco128/labels/train2017.cache' images and labels... 128 found, 0 missing, 2 empty, 0 corrupt: 100% 128/128 [00:00<?, ?it/s]
[34m[1mval: [0mCaching images (0.1GB ram): 100% 128/128 [00:00<00:00, 269.10it/s]
Plotting labels to runs/train/exp/labels.jpg...
[34m[1mAutoAnchor: [0m4.27 anchors/target, 0.994 Best Possible Recall (BPR). Current anchors are a good fit to dataset ✅
Image sizes 640 train, 640 val
Using 8 dataloader workers
Logging results to [1mruns/train/exp[0m
Starting training for 3 epochs...
Epoch gpu_mem box obj cls labels img_size
0/2 3.76G 0.04529 0.06712 0.01835 323 640: 100% 8/8 [00:04<00:00, 1.65it/s]
Class Images Labels P R mAP@.5 mAP@.5:.95: 100% 4/4 [00:00<00:00, 4.07it/s]
all 128 929 0.666 0.611 0.684 0.452
Epoch gpu_mem box obj cls labels img_size
1/2 4.79G 0.04244 0.06423 0.01611 236 640: 100% 8/8 [00:01<00:00, 7.60it/s]
Class Images Labels P R mAP@.5 mAP@.5:.95: 100% 4/4 [00:01<00:00, 3.90it/s]
all 128 929 0.746 0.626 0.722 0.481
Epoch gpu_mem box obj cls labels img_size
2/2 4.79G 0.04695 0.06875 0.0173 189 640: 100% 8/8 [00:00<00:00, 8.49it/s]
Class Images Labels P R mAP@.5 mAP@.5:.95: 100% 4/4 [00:00<00:00, 4.24it/s]
all 128 929 0.774 0.647 0.746 0.499
3 epochs completed in 0.003 hours.
Optimizer stripped from runs/train/exp/weights/last.pt, 14.9MB
Optimizer stripped from runs/train/exp/weights/best.pt, 14.9MB
Validating runs/train/exp/weights/best.pt...
Fusing layers...
Model summary: 213 layers, 7225885 parameters, 0 gradients, 16.4 GFLOPs
Class Images Labels P R mAP@.5 mAP@.5:.95: 100% 4/4 [00:03<00:00, 1.11it/s]
all 128 929 0.774 0.647 0.746 0.499
person 128 254 0.87 0.697 0.806 0.534
bicycle 128 6 0.759 0.528 0.725 0.444
car 128 46 0.774 0.413 0.554 0.239
motorcycle 128 5 0.791 1 0.962 0.595
airplane 128 6 0.981 1 0.995 0.689
bus 128 7 0.65 0.714 0.755 0.691
train 128 3 1 0.573 0.995 0.602
truck 128 12 0.613 0.333 0.489 0.263
boat 128 6 0.933 0.333 0.507 0.209
traffic light 128 14 0.76 0.228 0.367 0.209
stop sign 128 2 0.821 1 0.995 0.821
bench 128 9 0.824 0.526 0.676 0.31
bird 128 16 0.974 1 0.995 0.611
cat 128 4 0.859 1 0.995 0.772
dog 128 9 1 0.666 0.883 0.647
horse 128 2 0.84 1 0.995 0.622
elephant 128 17 0.926 0.882 0.93 0.716
bear 128 1 0.709 1 0.995 0.995
zebra 128 4 0.866 1 0.995 0.922
giraffe 128 9 0.777 0.778 0.891 0.705
backpack 128 6 0.894 0.5 0.753 0.294
umbrella 128 18 0.876 0.783 0.899 0.54
handbag 128 19 0.799 0.209 0.335 0.179
tie 128 7 0.782 0.714 0.787 0.478
suitcase 128 4 0.658 1 0.945 0.581
frisbee 128 5 0.726 0.8 0.76 0.701
skis 128 1 0.8 1 0.995 0.103
snowboard 128 7 0.815 0.714 0.852 0.574
sports ball 128 6 0.649 0.667 0.602 0.307
kite 128 10 0.7 0.47 0.546 0.206
baseball bat 128 4 1 0.497 0.544 0.182
baseball glove 128 7 0.598 0.429 0.47 0.31
skateboard 128 5 0.851 0.6 0.685 0.495
tennis racket 128 7 0.754 0.429 0.544 0.34
bottle 128 18 0.564 0.333 0.53 0.264
wine glass 128 16 0.715 0.875 0.907 0.528
cup 128 36 0.825 0.639 0.803 0.535
fork 128 6 1 0.329 0.5 0.384
knife 128 16 0.706 0.625 0.666 0.405
spoon 128 22 0.836 0.464 0.619 0.379
bowl 128 28 0.763 0.607 0.717 0.516
banana 128 1 0.886 1 0.995 0.399
sandwich 128 2 1 0 0.62 0.546
orange 128 4 1 0.75 0.995 0.622
broccoli 128 11 0.548 0.443 0.467 0.35
carrot 128 24 0.7 0.585 0.699 0.458
hot dog 128 2 0.502 1 0.995 0.995
pizza 128 5 0.813 1 0.962 0.747
donut 128 14 0.662 1 0.96 0.838
cake 128 4 0.868 1 0.995 0.822
chair 128 35 0.538 0.571 0.594 0.322
couch 128 6 0.924 0.667 0.828 0.538
potted plant 128 14 0.731 0.786 0.824 0.495
bed 128 3 0.736 0.333 0.83 0.425
dining table 128 13 0.624 0.259 0.494 0.336
toilet 128 2 0.79 1 0.995 0.846
tv 128 2 0.574 1 0.995 0.796
laptop 128 3 1 0 0.695 0.367
mouse 128 2 1 0 0.173 0.0864
remote 128 8 1 0.62 0.634 0.557
cell phone 128 8 0.612 0.397 0.437 0.221
microwave 128 3 0.741 1 0.995 0.766
oven 128 5 0.33 0.4 0.449 0.3
sink 128 6 0.444 0.333 0.331 0.231
refrigerator 128 5 0.561 0.8 0.798 0.546
book 128 29 0.635 0.276 0.355 0.164
clock 128 9 0.766 0.889 0.888 0.73
vase 128 2 0.303 1 0.995 0.895
scissors 128 1 1 0 0.332 0.0397
teddy bear 128 21 0.842 0.508 0.739 0.499
toothbrush 128 5 0.787 1 0.928 0.59
Results saved to [1mruns/train/exp[0m
ClearML Logging and Automation 🌟 NEW
ClearML is completely integrated into YOLOv5 to track your experimentation, manage dataset versions and even remotely execute training runs. To enable ClearML (check cells above):
pip install clearml- run
clearml-initto connect to a ClearML server (deploy your own open-source server, or use our free hosted server)
You'll get all the great expected features from an experiment manager: live updates, model upload, experiment comparison etc. but ClearML also tracks uncommitted changes and installed packages for example. Thanks to that ClearML Tasks (which is what we call experiments) are also reproducible on different machines! With only 1 extra line, we can schedule a YOLOv5 training task on a queue to be executed by any number of ClearML Agents (workers).
You can use ClearML Data to version your dataset and then pass it to YOLOv5 simply using its unique ID. This will help you keep track of your data without adding extra hassle. Explore the ClearML Tutorial for details!
Weights & Biases Logging
Weights & Biases (W&B) is integrated with YOLOv5 for real-time visualization and cloud logging of training runs. This allows for better run comparison and introspection, as well improved visibility and collaboration for teams. To enable W&B pip install wandb, and then train normally (you will be guided through setup on first use).
During training you will see live updates at https://wandb.ai/home, and you can create and share detailed Reports of your results. For more information see the YOLOv5 Weights & Biases Tutorial.
Local Logging
Training results are automatically logged with Tensorboard and CSV loggers to runs/train, with a new experiment directory created for each new training as runs/train/exp2, runs/train/exp3, etc.
This directory contains train and val statistics, mosaics, labels, predictions and augmentated mosaics, as well as metrics and charts including precision-recall (PR) curves and confusion matrices.
Environments
YOLOv5 may be run in any of the following up-to-date verified environments (with all dependencies including CUDA/CUDNN, Python and PyTorch preinstalled):
- Google Colab and Kaggle notebooks with free GPU:
- Google Cloud Deep Learning VM. See GCP Quickstart Guide
- Amazon Deep Learning AMI. See AWS Quickstart Guide
- Docker Image. See Docker Quickstart Guide
Status
If this badge is green, all YOLOv5 GitHub Actions Continuous Integration (CI) tests are currently passing. CI tests verify correct operation of YOLOv5 training (train.py), testing (val.py), inference (detect.py) and export (export.py) on macOS, Windows, and Ubuntu every 24 hours and on every commit.
In [ ]:
import torch
# PyTorch Hub Model
model = torch.hub.load('ultralytics/yolov5', 'yolov5s') # or yolov5n - yolov5x6, custom
# Images
img = 'https://ultralytics.com/images/zidane.jpg' # or file, Path, PIL, OpenCV, numpy, list
# Inference
results = model(img)
# Results
results.print() # or .show(), .save(), .crop(), .pandas(), etc.In [ ]:
# YOLOv5 CI
%%shell
rm -rf runs # remove runs/
m=yolov5n # official weights
b=runs/train/exp/weights/best # best.pt checkpoint
python train.py --imgsz 64 --batch 32 --weights $m.pt --cfg $m.yaml --epochs 1 --device 0 # train
for d in 0 cpu; do # devices
for w in $m $b; do # weights
python val.py --imgsz 64 --batch 32 --weights $w.pt --device $d # val
python detect.py --imgsz 64 --weights $w.pt --device $d # detect
done
done
python hubconf.py --model $m # hub
python models/tf.py --weights $m.pt # build TF model
python models/yolo.py --cfg $m.yaml # build PyTorch model
python export.py --weights $m.pt --img 64 --include torchscript # exportIn [ ]:
# Reproduce
for x in (f'yolov5{x}' for x in 'nsmlx'):
!python val.py --weights {x}.pt --data coco.yaml --img 640 --task speed # speed
!python val.py --weights {x}.pt --data coco.yaml --img 640 --conf 0.001 --iou 0.65 # mAPIn [ ]:
# Profile
from utils.torch_utils import profile
m1 = lambda x: x * torch.sigmoid(x)
m2 = torch.nn.SiLU()
results = profile(input=torch.randn(16, 3, 640, 640), ops=[m1, m2], n=100)In [ ]:
# VOC
for b, m in zip([64, 64, 64, 32, 16], [f'yolov5{x}' for x in 'nsmlx']): # batch, model
!python train.py --batch {b} --weights {m}.pt --data VOC.yaml --epochs 50 --img 512 --hyp hyp.VOC.yaml --project VOC --name {m} --cacheIn [ ]:
# Classification train
for m in [*(f'yolov5{x}-cls.pt' for x in 'nsmlx'), 'resnet50.pt', 'resnet101.pt', 'efficientnet_b0.pt', 'efficientnet_b1.pt']:
for d in 'mnist', 'fashion-mnist', 'cifar10', 'cifar100', 'imagenette160', 'imagenette320', 'imagenette', 'imagewoof160', 'imagewoof320', 'imagewoof':
!python classify/train.py --model {m} --data {d} --epochs 10 --project YOLOv5-cls --name {m}-{d}In [ ]:
# Classification val
!bash data/scripts/get_imagenet.sh --val # download ImageNet val split (6.3G, 50000 images)
!python classify/val.py --weights yolov5m-cls.pt --data ../datasets/imagenet --img 224 # validateIn [ ]:
# Validate on COCO test. Zip results.json and submit to eval server at https://competitions.codalab.org/competitions/20794
!bash data/scripts/get_coco.sh --test # download COCO test-dev2017 (7GB - 40,000 images, test 20,000)
!python val.py --weights yolov5x.pt --data coco.yaml --img 640 --iou 0.65 --half --task testIn [ ]:
# TensorRT
!pip install -U nvidia-tensorrt --index-url https://pypi.ngc.nvidia.com # install
!python export.py --weights yolov5s.pt --include engine --imgsz 640 --device 0 # export
!python detect.py --weights yolov5s.engine --imgsz 640 --device 0 # inference