Inference Scenario Configuration File

Run the following command to convert the training model and parameters into the format that can be used by the inference service:

python3 scripts/infer_service_generation.py --model_params=model_service_param.json

**Table 1** Parameters in the model_service_param.json file
Parameter	Type	Default Value	Description
task_name	String	None	(Mandatory) Task name.
project_list	Array[Object]	None	(Mandatory) Project list.
custom_params	Object	None	(Optional) This field is custom.

**Table 2** Details about a project
Parameter	Type	Default Value	Description
scene	String	None	(Mandatory) Scenario name.
project_name	String	None	(Mandatory) Project name.
model_list	Array[Object]	None	(Mandatory) Model list.

**Table 3** Details about a single model
Parameter	Type	Default Value	Description
application	String	None	(Mandatory) Application scenarios: det: foreign object detection seg: image segmentation det_cls: component error/missing/reverse detection ocr: industrial OCR det_det: detection + detection det_seg: detection + image segmentation det_ocr: detection + industrial OCR tag_paste: tag defect detection wafer: wafer detection det_htp: high-performance detection
model_name	String	None	(Optional) Model name. However, it is mandatory in the wafer detection scenario.
result_path	String	None	(Mandatory) Path of the OM file and the train_params.config folder generated during training.
ref_path	String	None	(Optional) Path for storing the reference diagram. It is used in the wafer detection scenario.

**Table 4** custom_params parameter description
Parameter	Type	Default Value	Description
max_infer_timeout	String	5000	(Optional) Maximum inference delay.
FirstDetectionFilter	String	None	(Optional) Detection object frame area filtering. You can filter the object frame based on the area and confidence. Example: {'Type': 'Area', 'TopN': 0,'BottomN': 0,'MinArea': 0,'MaxArea': 0,'ConfThresh': 0.0}
tag_1	String	None	(Optional) Name of the tag to be detected. It is used in the tag defect detection scenario. 1 indicates the label sequence number.
tag_1_params	String	None	(Optional) Edge defect parameter configuration. It is used in the tag defect detection scenario. 1 indicates the label sequence number. Example: {'edge_defect_thres_all': 10,'shape': (1, 128, 512, 3),'physical_size': [70, 30]} edge_defect_thres_all: size of the edge defect area shape: input size of the classification model physical_size: physical size of the label

**Table 5** Parameters in **model_configs.json**
Parameter	Type	Default Value	Description
infer_scene	String	None	(Mandatory) Name of an inference scenario.
ProjectName	String	None	(Mandatory) Project name.
TaskName	String	None	(Mandatory) Task name.
config_path	String	None	(Optional) Configuration file path, which is used in semiconductor detection scenarios.
pipeline_path	String	None	(Mandatory) Pipeline file path.
default_device_id	int	0	(Mandatory) ID of the running processor.
custom_params	Dict	None	(Optional) User-defined parameter.

A compilation example of model_configs.json is as follows:

{
    "model_configs": [
        {
            "infer_scene": "wafer",
            "ProjectName": "project_wafer",
            "TaskName": "assm1-2",
            "config_path":"config/models/project_jsx/assm1-2/1/aoi_ai_config.json",
            "pipeline_path": "config/models/project_jsx/assm1-2/1/aoi_ai_sdk.pipeline",
            "default_device_id": 0,
            "custom_params": {
                "maximum_tag_num": "100"
            }
        }
    ]
}

**Table 6** Parameters in **aoi_ai_config.json** for semiconductor detection
Parameter	Type	Default Value	Description
station_names	Dict	None	(Mandatory) machine name : ID. Currently, only two types of machines are supported: "RudolphTechnologies": 0, "Camtek": 1
ref_key_exclude	List	None	(Optional) ID of the processor layer that requires center cropping.
labels	List	None	(Mandatory) Supported processor process.
convert_map	Dict	None	(Mandatory) The format is as follows: "convert_map": { "Processor ID": { "Processor process": { "Processor layer ID": ["M1"], "classes": ["A", "B"] } }
reg_range_lut	Dict	None	(Mandatory) The format is as follows: "reg_range_lut" : { "Processor ID and layer ID": [ [[Value range on the X-axis of machine R],[Value range on the X-axis of machine C]], [[Value range on the Y-axis of machine R],[Value range on the Y-axis of machine C]], registration threshold, original die width on the workstation, original die height on the workstation, width of processor grayscale reference, height of processor grayscale reference, ratio of the second microscope to the first microscope ] } For details, see Table 7.
preprocess.blur_threds	String	7	(Mandatory) Threshold for fuzzy filtering.
model_infer.model_num	int	4	(Mandatory) Number of inference models.
postprocess.reject_prob_thred_and_version.reject_prob_thred	Float	0.1	(Mandatory) Inference rejection threshold.
postprocess.reject_prob_thred_and_version.version	String	None	(Mandatory) Version number of an inference model.

**Table 7** Parameters in **reg_range_lut**
Index	Type	Value Range	Description
0	list	Two integer values within [0,100000]	Value range on the X-axis.
1	list	Two int values in the range of [0, 100000]	Value range on the Y-axis.
2	float	[0, 1]	Registration threshold.
3	float	(0, 100000)	Width of a die.
4	float	(0, 100000)	Height of a die.
5	float	(0, 100000)	Grayscale image width.
6	float	(0, 100000)	Grayscale image height.
7	float	[0.1, 10]	Magnification ratio.

A compilation example of aoi_ai_config.json is as follows:

{
    "station_names": {
        "RudolphTechnologies": 0,
        "Camtek": 1
    },
    "ref_key_exclude": ["00101PI3"],
    "labels": ["FOI_FOP","BMP"],
    "convert_map": {
        "001": {
            "FOI_FOP": {
                "layers": ["M1"],
                "classes": ["A","B","E","F","L","M","MS"]
            },
            "BMP": {
                "layers": [],
                "classes": [
                    ["A","B","D","E"],
                    ["C","F"]
                ]
            }
        }
    },
    "reg_range_lut": {
        "00101M1": [
            [[40,130],[57,97]],[[20,110],[71,111]],0.5,17752.80126254142,18257.015024389606,10140,10431,2
        ]
    },
    "preprocess": {
        "blur_threds": {
            "default": 7,
            "00101M1": 7.95
        }
    },
    "model_infer": {
        "model_num": 4
    },
    "postprocess": {
        "reject_prob_thred_and_version": {
            "default": {
                "reject_prob_thred": 0,
                "version": "beta"
            },
            "00101M1": {
                "reject_prob_thred": 0.3,
                "version": "1.1.1"
            }
        }
    }
}

Parent topic: Deployment and Running