Text Detection (CTPN)

Training Parameters and Value Ranges

The following table describes the names, types, value ranges, default values, and descriptions of training parameters of the CTPN model.

**Table 1** Parameters for CTPN model training (**ctpn_mindspore**/**model_train.py**)
Parameter	Type	Value Range	Default Value	Description
--train_dataset_path	String	-	None	Path of the training dataset.
--pretrained_ckpt_path	String	-	"pre_trained_ckpt"	Path of the pre-trained model.
--train_output_path	String	-	"train_output_path"	Output path of the training result.
--device_id	Integer	0–7	0	ID of the NPU used for training.
--device_num	Integer	The value can only be 1. That is, only single-device training is supported.	1	Number of NPUs used for training.
--input_width	String	An integer multiple of 16 within (0,1280)	960	Input width for training.
--input_height	String	An integer multiple of 16 within (0,1280)	576	Input height for training.
--batch_size	Integer	The value can only be 1.	1	Number of image batches for training.
--init_lr	Float	(0,1)	0.0005	Training learning rate.
--epoch_size	Integer	[1,10000]	10	Number of training epochs.
--run_eval	Bool	True or False	True	Whether to perform evaluation during training.
--eval_start_epoch	Integer	[1,10000]	1	Sequence number of epochs from which evaluation is performed during training.
--save_best_ckpt	Bool	True or False	True	Whether to save the optimal CKPT file. (This parameter will be deleted in later versions.) Currently, when run_eval is set to True, the optimal CKPT file is saved by default.)
--eval_interval	Integer	[1,10000]	1	Number of epochs at which evaluation is performed.

Training Command Reference

The command for training the CTPN model is as follows:

python3 model_train.py --train_dataset_path=/path_to/images --train_output_path=./output_dir --pretrained_ckpt_path=/path_to/ckpt --epoch_size=20 --batch_size=1 --input_width=960 --input_height=576 --init_lr=0.0005 --device_num=1 --device_id=0 --run_eval=True

The model training process is random. Use the evaluated accuracy. The following shows the training accuracy of four images and twenty epochs when the preceding parameters are used.

Figure 1 Training accuracy

After the training is complete, the log information is displayed as follows:

Figure 2 Log information

After the training is complete, the .ckpt, .a310.om, .a310p.om, and .air model files are generated in the output directory specified by the --train_output_path parameter.

Evaluation Parameters and Value Ranges

The following table describes the names, types, value ranges, default values, and descriptions of evaluation parameters of the CTPN model.

**Table 2** Evaluation parameters (**ctpn_mindspore**/**model_eval.py**)
Parameter	Type	Value Range	Default Value	Description
--eval_dataset_path	String	-	""	Path of the evaluation dataset.
--eval_ckpt_path	String	-	""	Path of the trained model.
--eval_output_path	String	-	''/eval_output_path''	Output path of the evaluation results.
--device_id	Integer	0–7	0	ID of the NPU used for evaluation.
--text_iou_thresh	Float	-	0.2	IoU threshold for text connection.

Evaluation Command Reference

The command for evaluating the CTPN model is as follows:

python3 model_eval.py --eval_dataset_path= path_to/images --eval_ckpt_path= path_to/your_ckpt --eval_output_path=./eval_result --device_id=0

The CKPT output during training is used to evaluate the model accuracy. The following is an example.

Figure 3 Model evaluation accuracy

For the folder and file generated in the evaluation directory, the all_images\ng_images\ok_images folder stores the image evaluation results, and the statistics.csv file stores the corresponding accuracy results.

Figure 4 Evaluation directory

Parent topic: Training Component