How do I set the training parameters?

[amYaniii]

Hello author, I have just touched this task and tried to train my own dataset. How do you determine the training epochs I want to run in the train.py file? and According to the official guidelines, I seem to have run through the code, but why are all my indicators 0 during training, is it because I have too few training sessions? Here's the command for me to run train.py "python train.py --train-images-folder "F:\edgeai-yolov5-yolo-pose\sports\images\train2017" --prepared-train-labels F:\lightweight-human-pose-estimation.pytorch-master\prepared_train_annotation.pkl --val-labels F:\lightweight-human-pose-estimation.pytorch-master\val_subset.json --val-images-folder "F:\edgeai-yolov5-yolo-pose\sports\images\val2017" --from-mobilenet" I deleted this command"--checkpoint-path F:\lightweight-human-pose-estimation.pytorch-master\mobilenet_sgd_68.848.pth.tar",because it would give an error, but I didn't change the network After removing this command, this will work through, but why are so many iterations being trained and the validation metric is 0? Is it because I didn't use pre-training weights?

[amYaniii]

What is the relationship between the iterations in the train.py file and the training round epochs, how many epochs are generally trained, and how to set the iterations?

[Daniil-Osokin]

Hi! These are warnings, not errors, so model initialization seems to be ok. Usually 1 epoch = going through all samples in datasets. So usually number_of_iterations_per_epoch = number_of_samples_in_dataset / batch_size

[amYaniii]

This issue has been resolved, thanks for your reply.