yolo-v5 code
1. Environmental preparation
yolo-v5 is more like a project
The algorithm is similar to v4, but the details are different.
Pull code
Environmental requirements
Because v6 uses the mixed precision in torch1.6
2. Data preparation
The training uses the coco data set
The coco data set is very large and takes time to train. You can directly use your own data set to do it.
Suggested data sets that can be used
File organization format
When downloading, you need to select the format of the data.
Even if this format
Different data have different annotation formats. If the format is not correct, it will not be read.
Test training validation data set
data configuration file
nc is the number of categories
Here is the annotation data
The lableme annotation tool marks the json format, and there are also tools to convert it into the yolo tit format.
What is the role of cache files? Create a cache and repeat the training faster next time.
3. Preparation of pre-training model
Does this refer to the structure of the network or the weight file of the trained model?
Different models have different effects
4. Operation
A test operation can be performed
You can also perform a training operation
4.1 Training operations
exlips can set training parameters here
The folder where the configuration file is located
During training, place the model under the root directory
Training can be performed on a single machine, a single card, or on multiple machines and multiple cards.
run folder training results, log information
This can download all model files
4.2 Execute the detection process
Parameters also need to be specified
Completed formats are provided
View execution results
Being able to run through the examples here shows that there are no problems with the pre-trained network model, environment configuration, and code configuration.
If it cannot be executed, you need to check whether there is something wrong with the environment configuration.
data data (open annotation data format)
model has a model file and a model configuration file
weight model