The trainval_percent of win10 tensorflow ssd represents the proportion of the cross-validation set to the total picture, and train_percent is determined by the proportion of the training set to the cross-validation set

In the learning process of win10 tensorflow ssd, I can see from other places: trainval_percent represents the proportion of the cross-validation set to the total pictures, and train_percent is the proportion of the training set to the cross-validation set. That is, how to determine this parameter in the pascalvoc_2007.py file in tensorflow ssd, the parameters are as follows:

SPLITS_TO_SIZES = {
    'train': 5011,
    'test': 4952,
}

Reference comments:

1 Reference: https://www.cnblogs.com/pacino12134/p/10415157.html

Four txt files in ImageSets\Main

Create a new folder in ImageSets, name it Main, and generate four txt files in the Main folder, namely:

test.txt is the test set
train.txt is the training set
val.txt is the validation set
trainval.txt is the training and validation set

In VOC2007, trainval is about 50% of the entire data set, and test is about 50% of the entire data set; train is about 50% of trainval, and val is about 50% of trainval.

The content in the txt file is the name of the sample picture (without suffix), and the format is as follows:

According to the generated xml, make trainval.txt in the VOC2007 dataset; train.txt; test.txt; val.txt
trainval accounts for 50% of the total dataset, test accounts for 50% of the total dataset; train accounts for 50% of the trainval , val accounts for 50% of trainval;
the above percentage can be modified according to your own data set, if the data set is relatively small, test and val can be less