how to train my own dataset?

[desperate1998]

I hava run your models successfully,but I want to know how to train model on my own dataset and what is the format of data. Thanks!

[job333]

同问

[Radishzinski]

我尝试用自己的coco格式数据训练网络，但是出现了cfg.LR.TRAIN.BASE_LR问题

[lijun857]

请问你们复现这个论文的结果和作者的差距大嘛

[akshayamadhuri]

@Radishzinski Check the learning rate used in the original paper or implementation: If you are using a pre-existing model architecture, it may be helpful to check the learning rate used in the original paper or implementation. This can give you a good starting point for determining an appropriate learning rate for your own dataset.

Use a learning rate schedule: Rather than specifying a single value for cfg.LR.TRAIN.BASE_LR, you can use a learning rate schedule that gradually decreases the learning rate over time. This can help the model converge more smoothly and avoid overfitting. One popular learning rate schedule is the "step" schedule, which decreases the learning rate by a factor of 10 after a certain number of epochs.

Experiment with different values: If you are still having issues with cfg.LR.TRAIN.BASE_LR, you can try experimenting with different values to see what works best for your specific dataset and model architecture. Keep in mind that the optimal learning rate may vary depending on the number of GPUs you are using, the batch size, and other training parameters. It may take some trial and error to find the optimal value.