PaddleX 3d Task Model Configuration File Parameters Explanation¶
Global¶
| Parameter Name | Data Type | Description | Default Value |
|---|---|---|---|
| model | str | Specifies the model name | Model name specified in the YAML file |
| mode | str | Specifies the mode (check_dataset/train/evaluate/export/predict) | check_dataset |
| dataset_dir | str | Path to the dataset | Dataset path specified in the YAML file |
| device | str | Specifies the device to use | Device ID specified in the YAML file |
| output | str | Output path | "output" |
| load_cam_from | str | Pre-trained parameter path of cam branch | The pre-trained parameter path of the cam branch specified in the YAML file |
| load_lidar_from | str | Pre-trained parameter path of lidar branch | The pre-trained parameter path of the lidar branch specified in the YAML file |
| datart_prefix | bool | Whether to add the root path to the dataset_dir | True |
| version | str | Dataset version | "mini" |
CheckDataset¶
| Parameter Name | Data Type | Description | Default Value |
|---|---|---|---|
| convert.enable | bool | Whether to convert the dataset format | False (Currently not supported) |
| split.enable | bool | Whether to re-split the dataset | False (Currently not supported) |
Train¶
Common Parameters for 3d Tasks¶
| Parameter Name | Data Type | Description | Default Value |
|---|---|---|---|
| epochs_iters | int | The number of times the model repeats learning the training data | Number of iterations specified in the YAML file |
| batch_size | int | Batch size | Batch size specified in the YAML file |
| learning_rate | float | Initial learning rate | Initial learning rate specified in the YAML file |
| warmup_steps | str | The number of steps performed in the warm-up training | The number of warm-up steps specified in the YAML file |
Evaluate¶
| Parameter Name | Data Type | Description | Default Value |
|---|---|---|---|
| batch_size | int | Batch size | Batch size specified in the YAML file |
| weight_path | str | Evaluation model path | Default local path from training output, when specified as None, indicates using official weights |
Export¶
| Parameter Name | Data Type | Description | Default Value |
|---|---|---|---|
| weight_path | str | Dynamic graph weight path for exporting the model | Default local path from training output, when specified as None, indicates using official weights |
Predict¶
| Parameter Name | Data Type | Description | Default Value |
|---|---|---|---|
| batch_size | int | Prediction batch size | The prediction batch size specified in the YAML file |
| model_dir | str | Path to the prediction model | The default local inference model path produced by training. When specified as None, it indicates the use of official weights |
| input | str | Path to the prediction input | The prediction input path specified in the YAML file |
| kernel_option.run_mode | str | Inference engine setting, such as: "paddle" | paddle |