Platform For AI:Object detection (easycv)

Prerequisites

OSS is activated, and Machine Learning Studio is authorized to access OSS. For more information, see Activate OSS and Grant required permissions to use Machine Learning Designer.

Limitations

• This component is available only in Designer.

• Only the DLC engine is supported.

Configure component parameters

• Input ports

  Input port	Type	Upstream component	Required
  Training data	Object Storage Service	Read OSS Data	No. If you do not connect this port, you can configure the Training data OSS path parameter on the Field settings tab.
  Validation data	Object Storage Service	Read OSS Data	No. If you do not connect this port, you can configure the Validation data OSS path parameter on the Field settings tab.
  Class list file	Object Storage Service	Read OSS Data	No. If you do not connect this port, you can configure the Class list file OSS path parameter on the Field settings tab.

• Component parameters

  Tab	Parameter	Required	Description	Default
  Field settings	Model type	Yes	The model architecture to use for training. Valid values: FCOS YOLOX YOLOv5 YOLOv7	YOLOX
  	OSS directory for training	No	The OSS directory to store the trained model. Example: examplebucket.oss-cn-shanghai-internal.aliyuncs.com/test/ckpt/. If you leave this parameter empty, the component uses the default workspace path.	None
  	Annotation file path for training set	No	If Data format is set to DetSourcePAI, provide a file with a .manifest extension. If Data format is set to COCO, provide a file with a .json extension. This parameter is not required if you provide training data through the input port. If both are configured, the input from the input port takes precedence.	None
  	Annotation file path for validation set	No	If Data format is set to DetSourcePAI, provide a file with a .manifest extension. If Data format is set to COCO, provide a file with a .json extension. This parameter is not required if you provide validation data through the input port. If both are configured, the input from the input port takes precedence.	None
  	Class list file OSS path	No	Specify the OSS path to the label list file, which must be a .txt file. This parameter is not required if you provide the class list file through the input port. If both are configured, the input from the input port takes precedence.	None
  	Pre-trained model OSS path	No	The OSS path to your custom pre-trained model. If you do not configure this parameter, PAI uses a default pre-trained model.	None
  	Training data OSS path	Yes	This parameter appears only when Data format is set to COCO. Specify the path to training images in COCO format.	None
  	Validation data OSS path	Yes	This parameter appears only when Data format is set to COCO. Specify the path to validation images in COCO format.	None
  	Data format	Yes	The format of the source dataset. Valid values: COCO DetSourcePAI (YOLOv5 and YOLOv7 models support only DetSourcePAI datasets.)	DetSourcePAI
  Parameters	YOLOX model structure	Yes	This parameter appears only when Model type is set to YOLOX. You can select a structure from the drop-down list. Valid values: yolox-s yolox-m yolox-l yolox-x	yolox-s
  	Number of classes	Yes	The number of class labels in the dataset.	20
  	Image scale	Yes	The size of the image after resizing. Use a space to separate the height and width. Example: 320 320.	320 320
  	Optimizer	Yes	This parameter is available only when Model type is set to YOLOX. The optimization method for model training. Valid values: momentum adam	momentum
  	Initial learning rate	Yes	The initial learning rate.	0.01
  	Training batch size	Yes	The number of samples per training iteration.	8
  	Validation batch size	Yes	The number of samples per validation iteration.	8
  	Number of epochs	Yes	The total number of training epochs. Note For YOLOX models, the value of this parameter must be greater than the sum of the values of the Warmup epochs and Last no augmented lr epochs parameters.	20
  	Log loss frequency	No	The interval for printing the loss value. The default value is 200, which indicates that the loss value is printed every 200 training batches.	200
  	Validation frequency	No	The validation interval in epochs. The default value is 2, which indicates that validation is performed every two epochs.	2
  	Warmup epochs	No	This parameter applies only to YOLOX models.	5
  	Last no augmented lr epochs	No	This parameter applies only to YOLOX models.	5
  	Export model format	Yes	The format of the exported model. Valid values: raw jit (The jit format is not supported for YOLOv5 and YOLOv7 models.) onnx	raw
  	Checkpoint saving frequency	No	The interval (in epochs) for saving checkpoint files. A value of 1 indicates that a checkpoint is saved after each epoch.	1
  Tuning	GPU machine type	Yes	This component requires a GPU instance to run.	4
  	Enable FP16	No	Enables or disables the FP16 half-precision mode.	false
  	Execution mode	Yes	Only distributed training on DLC is supported.	Distributed DLC

• Output ports

  Output port	Type	Downstream component
  Output model	The trained model is stored in SavedModel format at the path specified in the OSS directory for training parameter on the Field settings tab.	image prediction

Example

For FCOS, YOLOX, YOLOv5, and YOLOv7, you can use the Object detection (easycv) component to build a pipeline as shown in the following figure.

In this example, configure the components as follows:

1. Use iTAG in PAI to label images. For more information, see Create a dataset for data labeling and Create a labeling job.

   1. After you create a dataset, select the Object Detection template under General Template when creating a labeling job.

   2. In the Object Detection Label Configuration section, select Single-label. In the Label Configuration section, enter all required label names to facilitate the labeling process.

   3. 

   4. After the labeling is complete, export the labeling results in the manifest format to Object Storage Service (OSS) so that subsequent components can use them.

   5. 

2. Use the Read OSS Data component to read the labeling results file (for example, xxx.manifest). Set the OSS Data Path parameter of the Read OSS Data component to the OSS path of the dataset that contains the labeling results. Example: oss://examplebucket.oss-cn-shanghai.aliyuncs.com/ev_demo/xxx.manifest.

3. Connect the training and validation data to the Object detection (easycv) component and configure its parameters. For more information, see Configure component parameters.

For FCOS and YOLOX, you can also use the image prediction component to perform offline inference. For more information, see image prediction.

Related documents

• After an object detection model is trained, you can connect an image prediction component downstream of the Object detection (easycv) component to make predictions with the model and evaluate its performance. For more information, see image prediction.

• For more information about Designer components, see Designer overview.

• Designer provides multiple preset algorithm components. You can select the appropriate component for your use case. For more information, see Designer component reference.