positive666 / yolo_research Goto Github PK

问题反馈：无法加载自己数据文件txt，无法生成.cache文件，在dataset.py中，第512行

解决方法：要把img_files改为im_files即可

您好，我想把您的解耦头移植到我的v5代码里，但是我把common.py里的DecoupledHead以及 yolo.py修改好后训练时报错是哪里还没改对吗：

作者大大，我从您的model/hub 文件和yolov5原作者的代码里现在mater版本的model/hub里加载yolov5-transformer.yaml，运行yolo.py都出现了报错：TypeError: new(): argument 'size' must be tuple of ints, but found element of type float at pos 2，这是什么原因呀？

❔Question

你好，我在尝试采用carafe算子进行训练时，报错如下:
yolo.py in parse_model
c2_ = make_divisible(c2 * gw, 8)
TypeError: unsupported operand type(s) for *: 'NoneType' and 'float'
请问该如何解决呢？是哪里没设置对吗?

Additional context

我把您的Decoupled Head 移植到到了我的代码里。修改了common.py 和yolo.py.我运行yolo.py测试网络是否可以正常forward，但出现了以下问题：

28 [19, 23, 27] 1 7337351 Decoupled_Detect [10, [[18, 14, 12, 26, 23, 20], [33, 24, 19, 44, 45, 34], [66, 49, 104, 73, 168, 136]], [216, 216, 416]]

<class 'NoneType'> None

Traceback (most recent call last):
File "/workspace/xdy/yolov5/models/yolo.py", line 689, in
model = Model(opt.cfg,opt.feature).to(device)
File "/workspace/xdy/yolov5/models/yolo.py", line 332, in init
m.stride = torch.tensor([s / x.shape[-2] for x in self.forward(torch.zeros(1, ch, s, s))]) # forward
File "/workspace/xdy/yolov5/models/yolo.py", line 355, in forward
return self._forward_once(x, profile, visualize) # single-scale inference, train
File "/workspace/xdy/yolov5/models/yolo.py", line 378, in _forward_once
x = m(x) # run
File "/root/miniconda3/envs/torch17/lib/python3.7/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
result = self.forward(*input, **kwargs)
File "/workspace/xdy/yolov5/models/yolo.py", line 57, in forward
bs, _, ny, nx = x[i].shape # x(bs,255,20,20) to x(bs,3,20,20,85)
AttributeError: 'NoneType' object has no attribute 'shape'

class Decoupled_Detect中的forward 的 x[i]是个空值 用原来的detect倒是正常，求作者大大指教

❔Question

Additional context

Before submitting a bug report, please be aware that your issue must be reproducible with all of the following, otherwise it is non-actionable, and we can not help you:

• Current repo: run git fetch && git status -uno to check and git pull to update repo
• Common dataset: coco.yaml or coco128.yaml
• Common environment: Colab, Google Cloud, or Docker image. See https://github.com/ultralytics/yolov5#environments

If this is a custom dataset/training question you must include your train*.jpg, test*.jpg and results.png figures, or we can not help you. You can generate these with utils.plot_results().

🐛 Bug

A clear and concise description of what the bug is.

To Reproduce (REQUIRED)

Input:

import torch

a = torch.tensor([5])
c = a / 0

Output:

Traceback (most recent call last):
  File "/Users/glennjocher/opt/anaconda3/envs/env1/lib/python3.7/site-packages/IPython/core/interactiveshell.py", line 3331, in run_code
    exec(code_obj, self.user_global_ns, self.user_ns)
  File "<ipython-input-5-be04c762b799>", line 5, in <module>
    c = a / 0
RuntimeError: ZeroDivisionError

Expected behavior

A clear and concise description of what you expected to happen.

Environment

If applicable, add screenshots to help explain your problem.

• OS: [e.g. Ubuntu]
• GPU [e.g. 2080 Ti]

Additional context

Add any other context about the problem here.

❔Question

File "D:\Python\yolov5\utils\datasets.py", line 420, in init
cache, exists = np.load(cache_path, allow_pickle=True).item(), True # load dict

Additional context

作者您好，请问仓库拉取后跑SwinV2.yaml为什么报错啊,是目前没有上传完整吗？而且我看到计算量很大，请问作者您试过将SwinTransformer作为类似C3卷积块的算子替换C3吗

❔Question

Additional context

我想问一下用您的V2版本的yaml去做训练，发现模型比较难训练，这种的话怎么去解决

❔Question

将x6模型文件的检测层，替换成ASFF_Detect报错，请问现在ASFF还不支持P6系列的模型吗？

Additional context

❔Question

您好，我采用您的swin-head和decouple的yaml，二者结合起来，然后在多个服务器上训练，大概在100epch上出现结果全为0的状态

Additional context

❔Question

我发现博主大大再添加mhsa时后面并未加BN和silu层，这是有什么深意吗？

Additional context

在官方yolov5-v6.0中的models/hub文件夹下，作者提供了一个yolov5-bifpn.yaml模型配置文件。我看了下，官方作者只在head的检测中等目标（medium）的检测头加入了bifpn，请问作者您知道怎么在小目标检测头也添加bifpn吗？如果您有相关的yaml文件，可以提供一下吗？谢谢
我问的配置文件的链接：https://github.com/ultralytics/yolov5/blob/v6.0/models/hub/yolov5-bifpn.yaml

❔Question

可以具体讲一下代码里解耦头具体的方法过程吗

Additional context

很清楚啊 就是在head部分加两个分支的卷积组 一个分支：分类 ，另一个分支：回归和前景置信度 然后在CAT回去 就可以了 类似于YOLOX，里面可以作卷积核微调 因为FLOPs明显增大了 后续会在CSDN上作详细说明

❔Question

博主大大您好，我在使用您的decoupled head 进行实验，采用yolov5s_decoupled.yamy，训练了65个epoch（batch size = 4，其余参数未修改），在visdrone数据集上验证集的map0.5只有0.143，是我什么地方有缺漏忘记调整了吗？

Additional context

附上我的实验结果如下
Epoch gpu_mem box obj cls labels img_size
64/64 1.19G 0.09373 0.1588 0.02641 388 640: 100%|██████████| 1618/1618 [02:45<00:00, 9.75it/s]
Class Images Labels P R [email protected] [email protected]:.95: 100%|██████████| 69/69 [00:08<00:00, 7.75it/s]
all 548 38759 0.282 0.215 0.143 0.0459

def channel_shuffle(self, x, groups=2):
#RESHAPE----->transpose------->Flatten
B, C, H, W = x.shape
out = x.view(B, groups, C // groups, H, W).permute(0, 2, 1, 3, 4).contiguous()
out=out.view(B, C, H, W)
return out

GAM:
x_spatial_att = self.spatial_attention(x).sigmoid()
x_spatial_att=channel_shuffle(x_spatial_att,4)
out = x * x_spatial_att
return out

🐛 Bug

x_spatial_att -> self , x -> int

去掉self能跑起来

你好，在实验中使用C3模块是可以的，使用其他的融合的C3模块就报这样的错误：TypeError: empty(): argument 'size' must be tuple of ints, but found element of type float at pos 2 这是什么原因呢

❔Question

感觉BoT好像不如C3块和C3TR块，可以减少参数量但会拉低一部分精度啊

Additional context

❔Question

Additional context

报错：AttributeError: Can't get attribute 'SPPF' on <module 'models.common' from 'C:\Users\86153\yolov5-mosaic9\models\common.py'>

❔Question

Additional context

博主您好，现在swinV2已经开源了，您的SWINV2的backbone如何去使用官方的V2的预训练权重呢

❔Question

我需要一个实时的部署要求，按照操作，编译报错了。我调试了很久，希望您能更新添加一下使用说明

Additional context

def channel_shuffle(self, x, groups=2):
#RESHAPE----->transpose------->Flatten
B, C, H, W = x.shape
out = x.view(B, groups, C // groups, H, W).permute(0, 2, 1, 3, 4).contiguous()
out=out.view(B, C, H, W)
return out

GAM:
x_spatial_att = self.spatial_attention(x).sigmoid()
x_spatial_att=channel_shuffle(x_spatial_att,4)
out = x * x_spatial_att
return out

🐛 Bug

x_spatial_att -> self , x -> int

去掉self能跑起来

Before submitting a bug report, please be aware that your issue must be reproducible with all of the following, otherwise it is non-actionable, and we can not help you:

• Current repo: run git fetch && git status -uno to check and git pull to update repo
• Common dataset: coco.yaml or coco128.yaml
• Common environment: Colab, Google Cloud, or Docker image. See https://github.com/ultralytics/yolov5#environments

If this is a custom dataset/training question you must include your train*.jpg, test*.jpg and results.png figures, or we can not help you. You can generate these with utils.plot_results().

🐛 Bug

A clear and concise description of what the bug is.

To Reproduce (REQUIRED)

Input:

import torch

a = torch.tensor([5])
c = a / 0

Output:

Traceback (most recent call last):
  File "/Users/glennjocher/opt/anaconda3/envs/env1/lib/python3.7/site-packages/IPython/core/interactiveshell.py", line 3331, in run_code
    exec(code_obj, self.user_global_ns, self.user_ns)
  File "<ipython-input-5-be04c762b799>", line 5, in <module>
    c = a / 0
RuntimeError: ZeroDivisionError

Expected behavior

A clear and concise description of what you expected to happen.

Environment

If applicable, add screenshots to help explain your problem.

• OS: [e.g. Ubuntu]
• GPU [e.g. 2080 Ti]

Additional context

Add any other context about the problem here.

❔Question

Dear Writer, would you like to add DCN moudel into this YOLOv5 project? If you have created it, where can I find it?

Additional context

❔Question

请问完全按照项目文件设置，在自己的数据集上，使用yolov5s_swinv2_head.yaml文件，总是显示以下错误：
RuntimeError: Given groups=1, weight of size [1024, 128, 1, 1], expected input[1, 32, 32, 128] to have 128 channels, but got 32 channels instead
没有自己改动，貌似是输入维度不匹配问题，但是找不到根源在哪，求教。

Additional context

❔Question

Additional context

❔Question

我将yolos-asff改成yolom-asff报错这个问题，请教一下修改思路，是不是要在yolo.py和common里对asff进行特定的修改呢？
Given groups=1, weight of size [128, 512, 1, 1], expected input[1, 768, 8, 8] to have 512 channels, but got 768 channels instead
万分感谢您！！我的邮箱是[email protected] 如果方便可以邮箱交流下！谢谢！！

Additional context

❔Question

: 'Model' object has no attribute 'stride'这个问题您已经解决了，但是非常好奇，想请教您是如何解决这个问题的呢？这个问题的原因是什么？可以解答下嘛？谢谢您

Additional context

❔Question

Additional context

commmon.py文件出现4个bug,但是运行common.py文件还是可以的，于是用yolov5s_swinv2_head.yaml训练自己的数据集看看效果，epoch=100，p,r,map值都为0，gpu环境迭代几次就会报错结束训练。

❔Question

作者大大，我在训练完自己的数据集后，在detect.py里载入训练好的权重后运行，结果报错：AttributeError: 'DetectMultiBackend' object has no attribute 'fp16'，这是怎么回事呀

Additional context

❔Question

我看见用来解耦头之后和yolov5s比对，计算量增加一点，参数量降低了，请问一下FPS大概降低了多少呢

Additional context

[image]()

Before submitting a bug report, please be aware that your issue must be reproducible with all of the following, otherwise it is non-actionable, and we can not help you:

• Current repo: run git fetch && git status -uno to check and git pull to update repo
• Common dataset: coco.yaml or coco128.yaml
• Common environment: Colab, Google Cloud, or Docker image. See https://github.com/ultralytics/yolov5#environments

If this is a custom dataset/training question you must include your train*.jpg, test*.jpg and results.png figures, or we can not help you. You can generate these with utils.plot_results().

🐛 Bug

A clear and concise description of what the bug is.

To Reproduce (REQUIRED)

Input:

import torch

a = torch.tensor([5])
c = a / 0

Output:

Traceback (most recent call last):
  File "/Users/glennjocher/opt/anaconda3/envs/env1/lib/python3.7/site-packages/IPython/core/interactiveshell.py", line 3331, in run_code
    exec(code_obj, self.user_global_ns, self.user_ns)
  File "<ipython-input-5-be04c762b799>", line 5, in <module>
    c = a / 0
RuntimeError: ZeroDivisionError

Expected behavior

A clear and concise description of what you expected to happen.

Environment

If applicable, add screenshots to help explain your problem.

• OS: [e.g. Ubuntu]
• GPU [e.g. 2080 Ti]

Additional context

Add any other context about the problem here.

❔Question

Additional context您好，我想问一下，如果SWIN V2不好训练的话，您有采用拿SWIN 1.0版去修改吗？

AutoAnchor: 5.36 anchors/target, 1.000 Best Possible Recall (BPR). Current anchors are a good fit to dataset ✅
Traceback (most recent call last):
File "train.py", line 678, in
main(opt)
File "train.py", line 575, in main
train(opt.hyp, opt, device, callbacks)
File "train.py", line 305, in train
compute_loss = ComputeLoss(model) # init loss class
File "/home/wch/yolov5_research/utils/loss.py", line 107, in init
self.g2=h['fl_eiou_gamma'] # focal eiou loss gamma
KeyError: 'fl_eiou_gamma'
请问是啥情况

Before submitting a bug report, please be aware that your issue must be reproducible with all of the following, otherwise it is non-actionable, and we can not help you:

• Current repo: run git fetch && git status -uno to check and git pull to update repo
• Common dataset: coco.yaml or coco128.yaml
• Common environment: Colab, Google Cloud, or Docker image. See https://github.com/ultralytics/yolov5#environments

If this is a custom dataset/training question you must include your train*.jpg, test*.jpg and results.png figures, or we can not help you. You can generate these with utils.plot_results().

🐛 Bug

A clear and concise description of what the bug is.

To Reproduce (REQUIRED)

Input:

import torch

a = torch.tensor([5])
c = a / 0

Output:

Traceback (most recent call last):
  File "/Users/glennjocher/opt/anaconda3/envs/env1/lib/python3.7/site-packages/IPython/core/interactiveshell.py", line 3331, in run_code
    exec(code_obj, self.user_global_ns, self.user_ns)
  File "<ipython-input-5-be04c762b799>", line 5, in <module>
    c = a / 0
RuntimeError: ZeroDivisionError

Expected behavior

A clear and concise description of what you expected to happen.

Environment

If applicable, add screenshots to help explain your problem.

• OS: [e.g. Ubuntu]
• GPU [e.g. 2080 Ti]

Additional context

Add any other context about the problem here.

❔Question

Additional context

按您修改的解耦头改到自己代码还是报错，显示 self.m = nn.ModuleList(DecoupledHead(x,nc,anchors) for x in ch) if self.decoupled else nn.ModuleList(nn.Conv2d(x, self.no * self.na, 1) for x in ch) # output conv
TypeError: 'bool' object is not iterable

❔Question

Additional context

我在主干网络的C3模块后面添加SE-Net注意力机制，但是运行detect.py时报如下错误：RuntimeError: Sizes of tensors must match except in dimension 1. Expected size 28 but got size 27 for tensor number 1 in the list.

❔Question

Additional context

大佬，我加了ASFF_Detect,出现报错，ASFF AttributeError: 'Model' object has no attribute 'stride',请问怎么解决，用的是github上的代码

AttributeError: 'DecoupledHead' object has no attribute 'bias'

看着您的common和yolo文件改的，但不知道为啥报了这个错。

Epoch gpu_mem box obj cls labels img_size
0/199 4.87G 0.1106 0.01998 0 13 640: 100%|██████████| 567/567 [01:50<00:00, 5.14it/s]
Class Images Labels P R [email protected] [email protected]:.95: 0%| | 0/78 [00:00<?, ?
Traceback (most recent call last):
File "train.py", line 675, in
main(opt)
File "train.py", line 572, in main
train(opt.hyp, opt, device, callbacks)
File "train.py", line 396, in train
results, maps, _ = val.run(data_dict,
RuntimeError: expected scalar type Float but found Half
大佬，所有的swin都会出现这个问题是什么原因呢

❔Question

您好，之前跟您在博客上沟通过，看到您开了ISSUE，我按照您的改进思路，想在LOSS和标签匹配上做一下改进，但是我现在没有什么思路，您能再分享下或者一些改进这部分代码上的工作吗？

Additional context

❔Question

大佬，能不能讲解一下这个解耦头的作用呀？很想学习一下，麻烦您了

Additional context