jizhishutong / yolou Goto Github PK

YOLOv3、YOLOv4、YOLOv5、YOLOv5-Lite、YOLOv6-v1、YOLOv6-v2、YOLOv7、YOLOX、YOLOX-Lite、PP-YOLOE、PP-PicoDet-Plus、YOLO-Fastest v2、FastestDet、YOLOv5-SPD、TensorRT、NCNN、Tengine、OpenVINO

License: GNU General Public License v3.0

Shell 0.01% Python 3.41% Jupyter Notebook 96.58% Cython 0.01%

yolo ncnn openvino tensorrt tengine yolov5 yolov6 yoloe

yolou's Introduction

YOLOU：United, Study and easier to Deploy

The purpose of our creation of YOLOU is to better learn the algorithms of the YOLO series and pay tribute to our predecessors.

Here "U" means United, mainly to gather more algorithms about the YOLO series through this project, so that friends can better learn the knowledge of object detection. At the same time, in order to better apply AI technology, YOLOU will also join The corresponding Deploy technology will accelerate the implementation of the algorithms we have learned and realize the value.

At present, the YOLO series algorithms mainly included in YOLOU are:

Anchor-base: YOLOv3, YOLOv4, YOLOv5, YOLOv5-Lite, YOLOv7, YOLOv5-TPH, YOLO-Fastest v2, YOLO-LF, YOLO-SA, YOLOR, YOLOv5-SPD

Anchor-Free: YOLOv6-v1, YOLOv6-v2, YOLOX, YOLOE, YOLOX-Lite, FastestDet

Face-Detection: YOLOv5-Face, YOLOFace-v2

Segmentation: YOLOv5-Segment

KeyPoint: YOLOv7-Keypoint

Classfication: ResNet, DarkNet,......

Comparison of ablation experiment results

Model	size(pixels)	[email protected]	[email protected]:95	Parameters(M)	GFLOPs	TensorRT-FP32(b16) ms/fps	TensorRT-FP16(b16) ms/fps
YOLOv5n	640	45.7	28.0	1.9	4.5	0.95/1054.64	0.61/1631.64
YOLOv5s	640	56.8	37.4	7.2	16.5	1.7/586.8	0.84/1186.42
YOLOv5m	640	64.1	45.4	21.2	49.0	4.03/248.12	1.42/704.20
YOLOv5l	640	67.3	49.0	46.5	109.1
YOLOv5x	640	68.9	50.7	86.7	205.7
YOLOv6-T	640
YOLOv6-n	640
YOLOv6	640	58.4	39.8	20.4	28.8	3.06/326.93	1.27/789.51
YOLOv7	640	69.7	51.4	37.6	53.1	8.18/113.88	1.97/507.55
YOLOv7-X	640	71.2	53.7	71.3	95.1
YOLOv7-W6	1280	72.6	54.9
YOLOv7-E6	1280	73.5	56.0
YOLOv7-D6	1280	74.0	56.6
YOLOv7-E6E	1280	74.4	56.8
YOLOX-s	640	59.0	39.2	8.1	10.8	2.11/473.78	0.89/1127.67
YOLOX-m	640	63.8	44.5	23.3	31.2	4.94/202.43	1.58/632.48
YOLOX-l	640			54.1	77.7
YOLOX-x	640			104.5	156.2
v5-Lite-e	320	35.1		0.78	0.73	0.55/1816.10	0.49/2048.47
v5-Lite-s	416	42.0	25.2	1.64	1.66	0.72/1384.76	0.64/1567.36
v5-Lite-c	512	50.9	32.5	4.57	5.92	1.18/850.03	0.80/1244.20
v5-Lite-g	640	57.6	39.1	5.39	15.6	1.85/540.90	1.09/916.69
X-Lite-e	320	36.4	21.2	2.53	1.58	0.65/1547.58	0.46/2156.38
X-Lite-s	416	Training…	Training…	3.36	2.90
X-Lite-c	512	Training…	Training…	6.25	5.92
X-Lite-g	640	58.3	40.7	7.30	12.91	2.15/465.19	1.01/990.69

How to use

Install

git clone https://github.com/jizhishutong/YOLOU
cd YOLOU
pip install -r requirements.txt

Training

python train_det.py --mode yolov6 --data coco.yaml --cfg yolov6.yaml --weights yolov6.pt --batch-size 32

Detect

python detect_det.py --source 0  # webcam
                            file.jpg  # image 
                            file.mp4  # video
                            path/  # directory
                            path/*.jpg  # glob
                            'https://youtu.be/NUsoVlDFqZg'  # YouTube
                            'rtsp://example.com/media.mp4'  # RTSP, RTMP, HTTP stream

Re-parameterization

See reparameterization.ipynb

Pose estimation

yolov7-w6-pose.pt

See keypoint.ipynb.

Detect Inference Result

Segmentation Inference Result

KeyPoint Inference Result

Face-Detect Inference Result

DataSet

train: ../coco/images/train2017/
val: ../coco/images/val2017/

├── images            # xx.jpg example
│   ├── train2017        
│   │   ├── 000001.jpg
│   │   ├── 000002.jpg
│   │   └── 000003.jpg
│   └── val2017         
│       ├── 100001.jpg
│       ├── 100002.jpg
│       └── 100003.jpg
└── labels             # xx.txt example      
    ├── train2017       
    │   ├── 000001.txt
    │   ├── 000002.txt
    │   └── 000003.txt
    └── val2017         
        ├── 100001.txt
        ├── 100002.txt
        └── 100003.txt

Export ONNX

Expand

python export.py --weights ./weights/yolov6/yolov6s.pt --include onnx

In order to facilitate the deployment and implementation of friends here, all models included in YOLOU have been processed to a certain extent, and their pre- and post-processing codes can be used in one set, because the format and output results of the ONNX files they export are consistent.

YOLOv5

YOLOv6

YOLOv7

YOLOX

YOLOv5-Lite

YOLOX-Lite

YOLO-Fastest v2

Acknowledgements

Expand

https://github.com/ultralytics/yolov5

https://github.com/WongKinYiu/yolor

https://github.com/ppogg/YOLOv5-Lite

https://github.com/WongKinYiu/yolov7

https://github.com/meituan/YOLOv6

https://github.com/ultralytics/yolov3

https://github.com/Megvii-BaseDetection/YOLOX

https://github.com/WongKinYiu/ScaledYOLOv4

https://github.com/WongKinYiu/PyTorch_YOLOv4

https://github.com/WongKinYiu/yolor

https://github.com/shouxieai/tensorRT_Pro

https://github.com/Tencent/ncnn

https://github.com/Gumpest/YOLOv5-Multibackbone-Compression

https://github.com/positive666/yolov5_research

https://github.com/cmdbug/YOLOv5_NCNN

https://github.com/OAID/Tengine

https://github.com/meituan/YOLOv6/releases/tag/0.2.0

https://github.com/PaddlePaddle/PaddleDetection

Citing YOLOU

Expand

If you use YOLOU in your research, please cite our work and give a star ⭐:

 @misc{yolou2022,
  title = { YOLOU：United, Study and easier to Deploy},
  author = {ChaucerG},
  year={2022}
}

For more information, Please follow the QR code below：

yolou's People

Contributors

Stargazers

Watchers

Forkers

yangzhegithub unclelld 939810161 liu-b-s shituo123456 cyhycy hllyzms af-74413592 huangzhenjie xinsuinizhuan nirvana6234 zhulei2016 lkcl0928 liqishan abangdd hao-ux fortefrankie wz940216 jeozhao deftruth jiangyuewu qingjiudd githubbingochen cdl-cdl-cdl xsmile-love wstchhwp hanxun0121 l-net-1992 xiaohui0225 jo-dean wjpstudy recapture2018 kim222333 tongbook orientonubo fangwudi wendyjj zwl995 g101400 tianlan6767 mandylove1993 hust-wayne ai-llm2 lilin19890401 yangkk2019 shnhrtkyk chenwanger autogyro shining-love pustar cv-det zhangliyuan97 qiu-pinggaizi ycbob ggscore yonghuicai passengerxuhongli c-shawn anminhhung insight-mk1 pocean2001 nemonameless tzhirong m-liu1987 laozhuang727 sunlinlin-aragon polaris-f jonygu woshiybc qimaoyang feiyumengmeng songjiahao-wq tecsai luoyizhi516 klonggan zp1018 dreamer121121 callmedanieldaniel quliulangle chen990627 jaciyu lifeng0718 jjqcat sfh123456 feiyumengmeng159 favoriteatewatermelon codingonion sinead-li bestsongc w-arrior ritesh1991 sleepwalker-niu jie311 guofenggitlearning achang146 elsie0908 shizhanhao mackenbaron skyrookieyu eisen99

yolou's Issues

百度网盘提取码是多少呢？

请问将onnx转成engine文件的程序有吗？

something wrong

Yolov6 is anchor-free，please correct the information.

使用opencv部署yolov3,v4,v5,v6, v7, yolox和nanodet目标检测全集，包含C++和Python两种版本的程序

yolov6 and yolox

When will other pretrained models for yolox and other yaml profiles for yolov6 be released

yolov7权重是这个repo新训出来的吗？为什么精度比官方代码高了点？

yolov7-x 53.7？高出的也太多了？
另外Yolov5精度和官方一模一样，贸然问下是不是就是搬来的权重呢？
而yolox却低了官方很多，是实现上有什么不一样吗？还是没训好呢？

工程有点大，下载很慢

这个工程感觉有几百M，下载有点慢，建议权重文件最好别上传在这个里面啊

how can i get coco_kpts_128 dataset

if there is a download link?

Question about pretrain weights

Can the pre-trained model released by the official repository be used normally in this repository?

yolov5 -seg

感谢博主的开源，我想使用yolov5-seg进行实例分割请问数据格式是什么样的？

MIT License

Can I use your source for commercial purposes?

GIOU计算有问题

Megvii-BaseDetection/YOLOX#712

Inference with onnx

Hi jizhishutong, do you have a script to do inference with onnx, just to check if the model works normally. Thanks.

yolov7训练错误

使用yolov7训练数据时，出现了下列错误：
RuntimeError: result type Float can't be cast to the desired output type __int64
目前在网上还没有找到yolov7相关解决方案，我的运行环境是pytorch==1.12,cuda==11.3,python==3.7

百度网盘预训练模型的提取码是多少呢？

yolov7 推理精度异常

yolov7 推理的时候，同样的模型同样的图片，与官方的推理代码的结果相差很远。一个是0.5 官方有0.91

Do I need to reparameterize trained YOLOv7 model by myself before exporting?

If yes, is it possible to provide a reparameterization script (like reparameterization.ipynb in official yolov7 repo)？

Yolov5 error when training

outputs must be a Tensor or an iterable of Tensors

更简洁的端到端加速，支持YOLO系列算法

https://github.com/Linaom1214/TensorRT-For-YOLO-Series/blob/main/Examples.ipynb

yoloe.py中，channel_list写的是[64,128,256],但在后面的build_effidehead_layer中，写的是channel_list[6],还有[8],[10]，？？？？这给的数据不够全吧，还是写错了？？

How is '--swin_float' implemented?

不再维护了吗？

这个项目好长时间没有更新了，是不再维护了吗？

缺少 yolov4 系列的模型

缺少 yolov4 系列的模型，yolov4标准版、yolov4-tiny，以及Scaled-yolov4系列

Will it support PP-YOLO？

some issus！

File "/home/pc/dawn_work_space/yolou/YOLOU/models/yolox.py", line 210, in get_output_and_grid
reg_box[..., :2] = (reg_box[..., :2] + grid) * stride
RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:1 and cuda:0!

yolov7 train error

请问支持trt推理yolov7-pose吗

yolov6s.pt这个文件不能加载，提示出错啊

export: weights=['./weights/yolov6s.pt'], imgsz=[640, 640], batch_size=1, device=cpu, half=False, inplace=False, train=False, optimize=False, dynamic=True, simplify=True, opset=12, include=['onnx']
YOLOv5 2022-8-25 Python-3.10.5 torch-1.12.0+cpu CPU

Traceback (most recent call last):
File "E:\Code\YOLOU-main\export.py", line 209, in
main(opt)
File "E:\Code\YOLOU-main\export.py", line 204, in main
run(**vars(opt))
File "E:\Code\pythonCode\changeable_region_detection-main\venv\lib\site-packages\torch\autograd\grad_mode.py", line 27, in decorate_context
return func(*args, **kwargs)
File "E:\Code\YOLOU-main\export.py", line 133, in run
model = attempt_load(weights, device=device, inplace=True, fuse=True) # load FP32 model
File "E:\Code\YOLOU-main\models\experimental.py", line 228, in attempt_load
ckpt = torch.load(attempt_download(w), map_location='cpu') # load
File "E:\Code\pythonCode\changeable_region_detection-main\venv\lib\site-packages\torch\serialization.py", line 712, in load
return _load(opened_zipfile, map_location, pickle_module, **pickle_load_args)
File "E:\Code\pythonCode\changeable_region_detection-main\venv\lib\site-packages\torch\serialization.py", line 1049, in _load
result = unpickler.load()
File "E:\Code\pythonCode\changeable_region_detection-main\venv\lib\site-packages\torch\serialization.py", line 1042, in find_class
return super().find_class(mod_name, name)

ModuleNotFoundError: No module named 'yolov6'

进程已结束,退出代码1

Repvgg convert in export

Hi jizhishutong, I check the export onnx model of yolov6 and find that some rep module is not reparameterized. Wanna know how to reparameterize when exporting, thanks!

YOLOv6s网络结构有问题，且权重不能载入

关于val_det.py推理时参数不匹配

当我尝试运行val_det.py时,提示forward()出现了未知的参数val,然后查看代码

但是去models/common.py中查看DetectMultiBackend()却发现forward()中并没有val这个参数

所以这应该是个小bug或者是没有及时更新的地方,无论是在common.py的forward()中添加一个val参数或者是去掉推理时的val=True都可以解决这个问题.

How to train my dataset? there are some issues when load pretrained weights.

size mismatch for model.33.cls_preds.2.bias: copying a param with shape torch.Size([80]) from check point, the shape in current model is torch.Size([14]).

Can you upload the weights to release?

It is inconvenient to download them outside of China. Maybe it is a good idea to upload them in the release, where large files are supported.

Which models do you support for prediction?

Does detect.py only support the yolov5 model?

似乎没有关闭马赛克增强

if mode == 'yolox':
if epoch == epochs - no_aug_epochs:
LOGGER.info("--->No mosaic aug now!")
dataset.mosaic = False # close mosaic
LOGGER.info("--->Add additional L1 loss now!")

我自己单步调试的时候发现， dataset.mosaic 在运行过程中还是True

pretrained weight

How can I get a pretrained weight about YOLOX, will train the model from the beginning?Thank you!

yolov7 loss

something wrong with yolov7

Password to download pretrained models in missing

Hi,

What is the password to access the Baidu Drive ?

yolo pose

You can find yolo with pose form this. https://github.com/BingfengYan/yolo_pose.
train+test+to tensorRT

第一个epoch开始训练时报错，请问这是什么问题？

Some questions about your experiment

Your experiment is based on COCO dataset? If it is true, some data is not consistent the official data.

Yolov7 error when training

When training yolov7, there comes
name 'DownC' is not defined.
Seems some modules not defined

yolov7训练出错

命令：python train.py --mode yolov7 --use_aux True --data garbage.yaml --cfg yolov7.yaml --weights weights/yolov7/yolov7.pt --batch-size 8
出现错误
**Logging results to runs/train/exp14
Starting training for 100 epochs...

 Epoch   gpu_mem       box       obj       cls    labels  img_size

0%| | 0/5 [00:01<?, ?it/s]
Traceback (most recent call last):
File "train.py", line 772, in
main(opt)
File "train.py", line 670, in main
train(opt.hyp, opt, device, callbacks)
File "train.py", line 448, in train
loss, loss_items = compute_loss(pred, targets.to(device), imgs) # init loss class
File "YOLOU/utils/loss.py", line 992, in call
bs_aux, as_aux_, gjs_aux, gis_aux, targets_aux, anchors_aux = self.build_targets2(p[:self.nl], targets, imgs)
File "YOLOU/utils/loss.py", line 1202, in build_targets2
indices, anch = self.find_5_positive(p, targets)
File "YOLOU/utils/loss.py", line 1388, in find_5_positive
indices.append((b, a, gj.clamp_(0, gain[3] - 1), gi.clamp_(0, gain[2] - 1))) # image, anchor, grid indices
RuntimeError: result type Float can't be cast to the desired output type long int
**

jizhishutong / yolou Goto Github PK

yolou's Introduction

YOLOU：United, Study and easier to Deploy

How to use

Install

Training

Detect

Re-parameterization

Pose estimation

Detect Inference Result

Segmentation Inference Result

KeyPoint Inference Result

Face-Detect Inference Result

Export ONNX

YOLOv5

YOLOv6

YOLOv7

YOLOX

YOLOv5-Lite

YOLOX-Lite

YOLO-Fastest v2

Acknowledgements

Citing YOLOU

For more information, Please follow the QR code below：

yolou's People

Contributors

Stargazers

Watchers

Forkers

yolou's Issues

Recommend Projects

Recommend Topics

Recommend Org