halotrouvaille / yolo-multi-backbones-attention Goto Github PK

Hello I am trying to train with a custom dataset and I am facing the following issue:

All images of my custom dataset is 448x320. I already made the changes in image size parameter, but this error still showing.

Could someone help me with this issue?

Can someone provide a new link to download the datasets? Or can I use the original datasets directly?

do we need to add code in forward in create_module?Thank you your help!

RT

How to train a model using custom data in YAML form like yolov5?

Traceback (most recent call last):
File "train.py", line 467, in
opt.cfg = list(glob.iglob('./**/' + opt.cfg, recursive=True))[0] # find file
IndexError: list index out of range
Why?

Hi!

Thanks for sharing your work.

I have one question.

How do you set the kernel size of ECA Layer?

“activation=linear” appears in cfg files, but there is no description in “models.py”. Can you tell me what to do?

Is it possible convert these model construction way from cfg to directly from pytorch API?

您好，我没发现CBAM Block, ECA Block，不知道作者能否指点我一下，谢谢！

Hi HaloTrouvaille,

Thank you for your marvelous work!

Could you make a tutorial in detail about how to train with custom datasets?

你好，关于cfg文件我有个小问题：我看到原始yolov3的cfg文件中第一个卷积核的数量是64，而你的ghostnet的cfg文件中第一个卷积核数量是16，而且基本上每一个卷积核的数量都比原始的要少，这是为什么呢？

想请问作者yolo+shufflenetv2在visdrone训练的结果如何，是否可以告知一下ii

如题
作者可以提供预训练权重吗

在吗作者 有一些问题可否问下。cfg文件中的内容完整吗？

骨干网络为VoVNet的cfg文件以及对应的model.py网络解析内容大佬有吗？？

你好，量化前后为什么保存的模型大小都是180MB，量化并没有减小模型的大小？

How I can train it on a custom dataset? Where to change classes and input size?

I only saw se,but no cbam,please tell me where is it,thank you!

Thank you for your great work.An error occurred while I was running the code.
python3 train.py --data data/oxfordhand.data --batch-size 1 --cfg cfg/yolov3-tiny-hand-se.cfg --epochs 100 -sr --s 0.001 --img-size 640 --weights ''
......
RuntimeError: Sizes of tensors must match except in dimension 2. Got 40 and 48
Can you give me some advice to solve this problem?

你好作者，想问下有没有bdk100的数据集网盘呀，分享的链接取消了，感谢！！

您好, 代码里量化时没有对BN层做处理, 参数仍然是fp32, 这个在部署时要怎么处理呢?

您好，您给出的visdrone数据集的训练结果就是您代码中的默认参数的结果吗

2020.9.26
(1) Add VoVNet as the backbone. The result is excellent.

What about this version of the cfg file? Thank you

您好，我在用yolov3-tiny训练自己的数据集时出现了以下错误，能帮忙看下么
RuntimeError: invalid argument 0: Sizes of tensors must match except in dimension 1. Got 40 and 48 in dimension 2 at /pytorch/aten/src/THC/generic/THCTensorMath.cu:71

Hi,
You have mentioned the number of parameters for Ghostnet + Yolov3 to be 23.49M. I wanted to confirm if it is correct or it should be 2.349M? The number of parameters seem to be too larger than the other models.

Thanks.

你换了ShuffleUnit之后，除了模型方面创建方面，其他地方有没有修改，我的微信17868380981，我有个疑问想和你讨论。