z1069614715 / objectdetection_script

一些关于目标检测的脚本的改进思路代码，详细请看readme.md

Hello,Welcome to my github!

• 📯 我的哔哩哔哩空间（Bilibili Video） https://space.bilibili.com/286900343
• 📚 我的CSDN博客（CSDN Blog） https://blog.csdn.net/qq_37706472

Project

1. 基于pytorch实现的图像分类框架.pytorch-classifier
2. 一些关于目标检测的脚本和改进思路代码.objectdetection_script
3. 使用pyqt5封装yolov5、yolov7.yolov7-pyqt
4. 基于YOLOV8的实时人脸检测模型.ultralytics-face
5. 基于YOLOV9集成众多创新改进.yolov9-improve

objectdetection_script's Issues

预测标签，以及样本不平均的过采样训练方法

魔导，有个问题困扰了很久，我想在coco数据集的基础上加上自己的数据集训练，但是会导致样本标签不平均的问题，能不能出一些权重采样，之类的小trick的视频？另外想问一下如果是预训练模型，比如yolov8s.pt，他已经在coco上训练了，那我在我自定义数据集上训练以后，如何同时也预测之前就学到的标签呢？比如仍然要识别person、bus诸如此类的标签，实验一直没试出来。

预测不出来不知道是不是模型把那些遗忘掉了，还是说需要冻结参数之类的操作，还是因为我配置项的问题，是不是需要修改模型的yaml文件​(好像没成功)

YOLOv5中加入可变形卷积后，为什么每次训练精度不一样，设置了随机种子，不加可变形卷积每次训练精度一样

yolov5-DCNv3，linux安装完后在import处报错

YOLOv7-PConv

PConv may be just useful in only 1 GPU, I run it in two GPUs, it doesn't work. So it can be resolved?

rt-detr-resnet50的预训练文件

博主我跑的u版的rt-detr-resnet50的网络结构是584层，GFLOPs是130.8，博主是修改了u版的代码吧，没有用ResNetlayer,换成了ConvNormLayer？使用resnet50的预训练文件出错了

大佬您好，请问您有遇到过pow() takes no keyword argumentspow这个问题吗？

高版本numpy中存在函数问题

colors = np.random.uniform(0, 255, size=(len(model_names), 3)).astype(np.int)
这句代码在高于1.2版本的numpy中会出现错误AttributeError: module 'numpy' has no attribute 'int'.
将np.int改为int可以解决问题

大佬，yolov5+nwd中w和h的计算代码是否存在问题呢？

yolov7 heat_map出现了一个bug

我的yolov7模型需要固定图片的大小为512，512，我更改以下代码：

图像有上下的黑边，这是一个情况

我尝试提前把图片处理好为512512，再运行代码
只要是512512的图片就什么输出都没有，进度条为0就终止了，就请问这个yolov7_heatmap有限制图片大小或者该怎么更改这种情况？

Aux OTA报错

tobj[b, a, gj, gi] = (1.0 - self.gr) + self.gr * iou.detach().clamp(0).type(tobj.dtype)  # iou ratio
RuntimeError: shape mismatch: value tensor of shape [71, 1] cannot be broadcast to indexing result of shape [71]

OTA报错

  File "/home/gem/utils/loss_OTA.py", line 291, in __call__
    tobj[b, a, gj, gi] = (1.0 - self.gr) + self.gr * iou.detach().clamp(0).type(tobj.dtype)  # iou ratio
RuntimeError: shape mismatch: value tensor of shape [57, 1] cannot be broadcast to indexing result of shape [57]

yolov8-track.py的问题

代码72行 tracker.update(mot_input, image_plot)报错，显示AssertionError: Unsupported 'dets' dimensions, valid number of dimensions is two。

YOLOV8图像分割

作图像分割的的同学可以把画预测框的地方138行注释掉就可以使用了

goldyolo中报错：TypeError: 'int' object is not iterable

RuntimeError: shape mismatch: value tensor of shape [41, 1] cannot be broadcast to indexing result of shape [41]

大佬，在添加OTA中，拉的版本是yolov5 7.0 按照视频步骤，还是有如下问题
Traceback (most recent call last):
File "/home//yolov5/train.py", line 634, in
main(opt)
File "/home//yolov5/train.py", line 528, in main
train(opt.hyp, opt, device, callbacks)
File "/home//yolov5/train.py", line 310, in train
loss, loss_items = compute_loss(pred, targets.to(device),imgs) # loss scaled by batch_size
File "/home//yolov5/utils/loss.py", line 300, in call
tobj[b, a, gj, gi] = (1.0 - self.gr) + self.gr * iou.detach().clamp(0).type(tobj.dtype) # iou ratio
RuntimeError: shape mismatch: value tensor of shape [41, 1] cannot be broadcast to indexing result of shape [41]

yolov7-NWD

Hi, in yolov7-NWD.py，“tbox” may be replaced with "seleted_tbox"?

yolov7-DSConv是不是代码不全啊，DSConv2D下面没有前向传播呀

up主可以看到嘛，还请作者看到回复一下

YOLOX热力图

请问up后续有计划添加YOLOX的热力图吗？

很棒的工作，感谢分享！

How to add EfficientViT in existing yolov5

How to include EfficientViT in existin yolov5,,,, pls help

iou代码疑惑

iou = bbox_iou(pbox, tbox[i], CIoU=True)
这部分后面为什么不用加.squeeze()呢

导现在yolov5-asf里的MAX_pool3d已经改掉了，好像变成nn.functional.max_pool3d，这个了，然后在kernel_size基础上多了个input

DCNv3加入yolov5，DCNv3.dcnv3_forward报错RuntimeError: Not implemented on the CPU

windows系统，进行了DCNv3的编译。按照DCNv3训练报错

NWD_based NMS

想请教一下up主，怎么修改nms为基于NWD的呢？原文里面有提到，但是没找到源代码

maybe a bug

https://github.com/z1069614715/objectdetection_script/blob/master/yolo-improve/yolov5-DecoupledHead.py
Line 70 "if m is Detect:" should be "if isinstance(m, Detect)".
Similarly for Line 76

老师您好，我将你复现的项目拉到程序中最后val.py报错了

Fusing layers...
Traceback (most recent call last):
File "val.py", line 399, in
main(opt)
File "val.py", line 372, in main
run(**vars(opt))
File "C:\Users\hoshi\miniconda3\envs\air\lib\site-packages\torch\autograd\grad_mode.py", line 28, in decorate_context
return func(*args, **kwargs)
File "val.py", line 141, in run
model = DetectMultiBackend(weights, device=device, dnn=dnn, data=data)
File "I:\AIR\models\common.py", line 492, in init
model = attempt_load(weights if isinstance(weights, list) else w, map_location=device)
File "I:\AIR\models\experimental.py", line 98, in attempt_load
model.append(ckpt.fuse().eval() if fuse else ckpt.eval()) # fused or un-fused model in eval mode
File "I:\AIR\models\yolo.py", line 282, in fuse
m.fuse_repvgg_block()
File "C:\Users\hoshi\miniconda3\envs\air\lib\site-packages\torch\nn\modules\module.py", line 1177, in getattr
raise AttributeError("'{}' object has no attribute '{}'".format(
我拉到的项目不是yolov5原作，而是yoloair里面，你方便帮我看看是个什么原因么

RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation

Transferred 348/349 items
0%| | 0/504 [00:00<?, ?it/s]
Traceback (most recent call last):
File "yolov5_heatmap.py", line 117, in
model(r'E:\test\envyolo5\yolov5C3\datasets\00194.jpg', 'result')
File "yolov5_heatmap.py", line 81, in call
score.backward(retain_graph=True)
File "E:\test\envyolo5\lib\site-packages\torch\tensor.py", line 245, in backward
torch.autograd.backward(self, gradient, retain_graph, create_graph, inputs=inputs)
File "E:\test\envyolo5\lib\site-packages\torch\autograd_init_.py", line 147, in backward
allow_unreachable=True, accumulate_grad=True) # allow_unreachable flag
RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.cuda.FloatTensor [1, 3, 20, 20, 7]], which is output 0 of SigmoidBackward, is at version 2; expected version 0 instead. Hint: enable anomaly detection to find the operation that failed to compute its gradient, with torch.autograd.set_detect_anomaly(True).

请问UP这个怎么破，谢谢

关于WIoU损失函数

elif scale:
      return getattr(WIoU_Scale, '_scaled_loss')(self), (1 - iou) * torch.exp((rho2 / c2)), iou
else:
       return iou, torch.exp((rho2 / c2)) # WIoU v1

您好，请问两个问题：1.为什么返回多个iou损失？2.iou的均值iou_mean在WIoU_Scale里是怎么计算的？

关于热力图保存的问题

为什么保存的热力图会有上下两条灰色边条

在CAM的论文中作者还设计了FRM，您能否复现并开源FRM代码吗？

在yolov5中加入了CloAttention注意力机制报错

我用的是YOLOv5s模型为基础，在backbone第九层加入了CloAttention，照着视频里的做法，修改了yolo.py文件，但是训练的时候报错：
"RuntimeError: shape '[28, 2, -1, 64, 24]' is invalid for input of size 286720"
错误发生在第77行：
“kv = to_kv(kv).view(b, 2, -1, self.dim_head, (h*w)//(self.window_size**2)).permute(1, 0, 2, 4, 3).contiguous() #(2 b m (H W) d)”
是维度错误，每一轮训练结束后的验证阶段开始的时候报错.

SequentialSelfAttention注意力机制来源论文

SequentialSelfAttention注意力机制来源论文似乎不对呢，和上面的Polor……是重复的。请问SequentialSelfAttention来源论文是哪一篇呢？

MHSA这个论文是否正确呢？

yolov7+DCNv3 训练时报错：RuntimeError: Not implemented on the CPU

`File "/home/undergraduate/anaconda3/envs/yolov7p/lib/python3.8/site-packages/torch/cuda/amp/autocast_mode.py", line 110, in decorate_fwd
return fwd(*args, **kwargs)
File "/media/kb541/ruanpuchao/yolov7_train/models/ops_dcnv3/functions/dcnv3_func.py", line 39, in forward
output = DCNv3.dcnv3_forward(

RuntimeError: Not implemented on the CPU`

我按照视频将DCNv3添加到了yolov7网络中，但是当我开始训练的时候，程序报错RuntimeError: Not implemented on the CPU，我的命令行输入为：python track.py --device 0,1 --epoch 100 --batch_size 32，按理说不会再cpu上运行模型的啊，为什么报这个错误呢？谢谢。

如何在yolov7-tiny.yaml中加入PConv

比如某一个卷积层，我把[-1, 1, Conv, [32, 3, 1, None, 1, nn.LeakyReLU(0.1)]]改为[-1, 1, PConv, [32]]报错张量对不上

errno13 permission denied

导师你好，yolov5-7.0测试时老报错，麻烦您看一下。
`Speed: 1.1ms pre-process, 16.7ms inference, 4.0ms NMS per image at shape (1, 3, 640, 640)

Evaluating pycocotools mAP... saving runs\val\exp17\best_predictions.json...
pycocotools unable to run: [Errno 13] Permission denied: 'C:\Users\admin\Desktop\yolov5-7.0'
Results saved to runs\val\exp17
loading annotations into memory...

Process finished with exit code 0`

S2Attention bug？

class S2Attention(nn.Module):
def init(self, channels=512):
super().init()
self.mlp1 = nn.Linear(channels, channels * 3)
self.mlp2 = nn.Linear(channels, channels)
self.split_attention = SplitAttention()

这一行：
self.split_attention = SplitAttention()
好像要改为：
self.split_attention = SplitAttention(channels)
否则会报“矩阵乘法”相关的错误

（即，class SplitAttention(nn.Module): 里的
hat_a = self.mlp2(self.gelu(self.mlp1(a))) # bs,kc
不运行？）

DCNV3安装成功的cuda,torch版本

可以发一下DCNV3安装成功的cuda,torch版本吗

DCNv4 + question

How can i get the code from yolov8 from Europe?

yolov7改进（加注意力机制）

请问可以更新添加CA.py注意力机制的代码吗？我在github找的CA.py的源代码按照你的方法添加一直报错（我是选择在Concat后面加注意力机制），报错如下：

Traceback (most recent call last):
File "/hone/hndx/11z/yolow7-main-sE_cA/models/yolo. py",1ine 830, in
model=Model(opt. cfg). to(device)
File"/home/hudx/11z/yolow7-main-sE CA/nmodels/yolo. py", line 528, in init.
self. model, self. save=parse. model(deepcopy(seLf. yamt), ch=[ ch])# model, savelist
File "/home/hndx/LLz/yolayZ-main-SE CA/modets/yotopy", line 806, in parse_model m=nn. Sequential([m( args) forin range(n)]) if n>1 else m(* args)# module
FiLe "/home/hnex/11z/yoLov7-main-SE CA/models/comon. py", Line 69, in __init __
seLf. att=CoordAtt(channel
TypeError: init-() missing 1 required positionat argument:' oup'

希望作者可以帮忙解答，谢谢

no "backbone" in yaml file in yolo_v8_heatmap

I receive an error when there is no "backbone" in the yolo_v8 yaml config. Would like to ask if you can provide example yaml config files?

/yolov8_heatmap.py in __init__(self, weight, cfg, device, method, layer, backward_type, conf_threshold, ratio)
    52         model_names = ckpt['model'].names
    53         csd = ckpt['model'].float().state_dict()  # checkpoint state_dict as FP32
--> 54         model = Model(cfg, ch=3, nc=len(model_names)).to(device)
    55         csd = intersect_dicts(csd, model.state_dict(), exclude=['anchor'])  # intersect
    56         model.load_state_dict(csd, strict=False)  # load

/opt/conda/lib/python3.8/site-packages/ultralytics/nn/tasks.py in __init__(self, cfg, ch, nc, verbose)
    174             LOGGER.info(f"Overriding model.yaml nc={self.yaml['nc']} with nc={nc}")
    175             self.yaml['nc'] = nc  # override yaml value
--> 176         self.model, self.save = parse_model(deepcopy(self.yaml), ch=ch, verbose=verbose)  # model, savelist
    177         self.names = {i: f'{i}' for i in range(self.yaml['nc'])}  # default names dict
    178         self.inplace = self.yaml.get('inplace', True)

/opt/conda/lib/python3.8/site-packages/ultralytics/nn/tasks.py in parse_model(d, ch, verbose)
    437     ch = [ch]
    438     layers, save, c2 = [], [], ch[-1]  # layers, savelist, ch out
--> 439     for i, (f, n, m, args) in enumerate(d['backbone'] + d['head']):  # from, number, module, args
    440         m = getattr(torch.nn, m[3:]) if 'nn.' in m else globals()[m]  # get module
    441         for j, a in enumerate(args):

KeyError: 'backbone'

已解决

          已解决

Originally posted by @chao-code in #19 (comment)

如何加入到v7内使用？

是将整个Biformer.py加入到v7内？ 还是将某一段加入到common.py内 yolo.py里注册？