Yolov6 training: various problems encountered in training your dataset

I downloaded it before YoloV6s.pt Ran down , Although there is false detection, the confidence is still very high , So I plan to run my own data set

My data set has only two types , When I load my dataset, the problem arises ....

problem 1：

First look at the official data set format ...

But he will always report you wrong .....

 

Find out why , Step by step ,, Fucking ...

He wrote too much osp.dirname(img_dir),,, Let's print it out ..

 

This must give you an error ,,, Because your path is wrong , So the changes are as follows ..

I want to see others ...

ok! No problem .... 

Look at the data set format ：

CiWaData|

                train:  ./image.jpg

                val: ./image.jpg

               labels|  

                        train:./image.txt

                        val: ./image.txt

Start training ........

problem 2：

My heart is tired. .... A lot of questions ,,, This version is the latest ,2022/06/29 just git Download the ...

Fortunately, there is a new version In contrast ..  No matter the First modify to the previous version

 ok Finally, it's the next step ...

problem 3：

In training Map by 0, The first round and the second round are still 0, The back decisively closed the training ..

The first round ：

The second round :

I NIMA ！！！！！！！！！！ It feels like a strange upgrade ,, In the end, it's all a problem ...... incredibly MAp yes 0   ........ Forget it, forget it