Training dataset : 588 images
Validation dataset : 76 images
The config file looks like:
> "train_data_loader": {
"type": "DataLoader", "args": { "batch_size": 1, "shuffle": true, "drop_last": true, "num_workers": 4, "pin_memory": true } }, "val_data_loader": { "type": "DataLoader", "args": { "batch_size": 1, "shuffle": false, "drop_last": false, "num_workers": 4, "pin_memory": true } }, "optimizer": { "type": "Adam", "args": { "lr": 0.007, "weight_decay": 0.15, "amsgrad": true } }, "lr_scheduler": { "type": "StepLR", "args": { "step_size": 30, "gamma": 0.1 } }, "trainer": { "epochs": 50, "gl_loss_lambda": 0.015, "crf_loss_lambda": 1, "cn_loss_lambda": 10000, "log_step_interval": 20, "val_step_interval": 9999,
What could be the best set of parameters or a descent learning rate to start with?
Currently i get poor results and due to early stopping it is not improved.
Any suggestions ?