yolov5

YOLOv5

Abstract

YOLOv5 is a family of object detection architectures and models pretrained on the COCO dataset, representing Ultralytics open-source research into future vision AI methods, incorporating lessons learned and best practices evolved over thousands of hours of research and development.

Requirements

mindspore	ascend driver	firmware	cann toolkit/kernel
2.3.1	24.1.RC2	7.3.0.1.231	8.0.RC2.beta1

Quick Start

Please refer to the GETTING_STARTED in MindYOLO for details.

Training

View More

- Distributed Training

It is easy to reproduce the reported results with the pre-defined training recipe. For distributed training on multiple Ascend 910 devices, please run

# distributed training on multiple GPU/Ascend devices
msrun --worker_num=8 --local_worker_num=8 --bind_core=True --log_dir=./yolov5_log python train.py --config ./configs/yolov5/yolov5n.yaml --device_target Ascend --is_parallel True

Similarly, you can train the model on multiple GPU devices with the above msrun command. Note: For more information about msrun configuration, please refer to here.

For detailed illustration of all hyper-parameters, please refer to config.py.

Note: As the global batch size (batch_size x num_devices) is an important hyper-parameter, it is recommended to keep the global batch size unchanged for reproduction or adjust the learning rate linearly to a new global batch size.

- Standalone Training

If you want to train or finetune the model on a smaller dataset without distributed training, please run:

# standalone training on a CPU/GPU/Ascend device
python train.py --config ./configs/yolov5/yolov5n.yaml --device_target Ascend

Validation and Test

To validate the accuracy of the trained model, you can use test.py and parse the checkpoint path with --weight.

python test.py --config ./configs/yolov5/yolov5n.yaml --device_target Ascend --weight /PATH/TO/WEIGHT.ckpt

To validate the accuracy of the trained model for resolution of 1280, you can use test.py and parse the checkpoint path with --weight and parse the image sizes with --img_size.

python test.py --config ./configs/yolov5/yolov5n6.yaml --device_target Ascend --weight /PATH/TO/WEIGHT.ckpt --img_size 1280

Performance

Experiments are tested on Ascend 910* with mindspore 2.3.1 graph mode.

model name	scale	cards	batch size	resolution	jit level	graph compile	ms/step	img/s	map	recipe	weight
YOLOv5	N	8	32	640x640	O2	377.81s	520.79	491.56	27.4%	yaml	weights
YOLOv5	S	8	32	640x640	O2	378.18s	526.49	486.30	37.6%	yaml	weights
YOLOv5	N6	8	32	1280x1280	O2	494.36s	1543.35	165.87	35.7%	yaml	weights
YOLOv5	S6	8	32	1280x1280	O2	524.91s	1514.98	168.98	44.4%	yaml	weights
YOLOv5	M6	8	32	1280x1280	O2	572.32s	1769.17	144.70	51.1%	yaml	weights
YOLOv5	L6	8	16	1280x1280	O2	800.34s	894.65	143.07	53.6%	yaml	weights
YOLOv5	X6	8	8	1280x1280	O2	995.73s	864.43	74.04	54.5%	yaml	weights

Experiments are tested on Ascend 910 with mindspore 2.3.1 graph mode.

model name	scale	cards	batch size	resolution	jit level	graph compile	ms/step	img/s	map	recipe	weight
YOLOv5	N	8	32	640x640	O2	233.25s	650.57	393.50	27.3%	yaml	weights
YOLOv5	S	8	32	640x640	O2	166.00s	650.14	393.76	37.6%	yaml	weights
YOLOv5	M	8	32	640x640	O2	256.51s	712.31	359.39	44.9%	yaml	weights
YOLOv5	L	8	32	640x640	O2	274.15s	723.35	353.91	48.5%	yaml	weights
YOLOv5	X	8	16	640x640	O2	436.18s	569.96	224.58	50.5%	yaml	weights

Notes

• map: Accuracy reported on the validation set.
• We refer to the official YOLOV5 to reproduce the P5 series model, and the differences are as follows: The single-device batch size is 32. This is different from the official codes.

References

[1] Jocher Glenn. YOLOv5 release v6.1. https://github.com/ultralytics/yolov5/releases/tag/v6.1, 2022.