I am training a Yolo Tiny Net model using the following .cfg:
[Common]
image_size: 448
batch_size: 16
num_classes: 21
max_objects_per_image: 21
[DataSet]
name: yolo.dataset.text_dataset.TextDataSet
path: VOC_Dataset/pascal_voc.txt
thread_num: 5
[Net]
name: yolo.net.yolo_tiny_net.YoloTinyNet
weight_decay: 0.0005
cell_size: 7
boxes_per_cell: 2
object_scale: 1
noobject_scale: 0.5
class_scale: 1
coord_scale: 5
[Solver]
name: yolo.solver.yolo_solver.YoloSolver
learning_rate: 0.000001
moment: 0.9
max_iterators: 1000000
pretrain_model_path: lenet_model_to_use.ckpt
train_dir: models/train
here lenet_model_to_use.ckpt
is a previously used trained with one label less than the actual training model (which I add a label and image so that the new one will recognize that new image)
I noticed that using the current learning_rate, the model training keeps on executing without the loss decreasing. My question is that I want to stop the training if the loss doesn’t decrease further after two successive steps.
What is the parameter name to add and which value and under which section to set in the .cfg file?