Can't converge Loss function in object detection API

[miladfa7]

I receive these classification loss spikes and no convergence on my own data.
python3 train.py --logtostderr –train_dir=training/ -pipeline_config_path=ssd_mobilenet_pretrain/pipeline.config

the output tensorboard --logdir training/

What is the reason for this problem?

[asmaamirkhan]

• 🤔 It is a normal result since step number is still 500
• 👀 You have to wait for at least 2000 steps (if your data is complex it will take more)