Instance Segmentation Module Development Tutorial¶

I. Overview¶

The instance segmentation module is a crucial component in computer vision systems, responsible for identifying and marking pixels that contain specific object instances in images or videos. The performance of this module directly impacts the accuracy and efficiency of the entire computer vision system. The instance segmentation module typically outputs pixel-level masks (masks) for each target instance, which are then passed as input to the object recognition module for subsequent processing.

II. Supported Model List¶

Model	Model Download Link	Mask AP	GPU Inference Time (ms) [Normal Mode / High-Performance Mode]	CPU Inference Time (ms) [Normal Mode / High-Performance Mode]	Model Size (M)	Description
Mask-RT-DETR-H	Inference Model/Trained Model	50.6	172.36 / 172.36	1615.75 / 1615.75	449.9 M	Mask-RT-DETR is an instance segmentation model based on RT-DETR. By adopting the high-performance PP-HGNetV2 as the backbone network and constructing a MaskHybridEncoder encoder, along with introducing IOU-aware Query Selection technology, it achieves state-of-the-art (SOTA) instance segmentation accuracy with the same inference time.
Mask-RT-DETR-L	Inference Model/Trained Model	45.7	88.18 / 88.18	1090.84 / 1090.84	113.6 M

❗ The above list features the 2 core models that the image classification module primarily supports. In total, this module supports 15 models. The complete list of models is as follows:

👉Model List Details

Model	Model Download Link	Mask AP	GPU Inference Time (ms) [Normal Mode / High-Performance Mode]	CPU Inference Time (ms) [Normal Mode / High-Performance Mode]	Model Size (M)	Description
Cascade-MaskRCNN-ResNet50-FPN	Inference Model/Trained Model	36.3	141.69 / 141.69	nan / nan	254.8 M	Cascade-MaskRCNN is an improved Mask RCNN instance segmentation model that utilizes multiple detectors in a cascade, optimizing segmentation results by leveraging different IOU thresholds to address the mismatch between detection and inference stages, thereby enhancing instance segmentation accuracy.
Cascade-MaskRCNN-ResNet50-vd-SSLDv2-FPN	Inference Model/Trained Model	39.1	147.62 / 147.62	nan / nan	254.7 M
Mask-RT-DETR-H	Inference Model/Trained Model	50.6	172.36 / 172.36	1615.75 / 1615.75	449.9 M	Mask-RT-DETR is an instance segmentation model based on RT-DETR. By adopting the high-performance PP-HGNetV2 as the backbone network and constructing a MaskHybridEncoder encoder, along with introducing IOU-aware Query Selection technology, it achieves state-of-the-art (SOTA) instance segmentation accuracy with the same inference time.
Mask-RT-DETR-L	Inference Model/Trained Model	45.7	88.18 / 88.18	1090.84 / 1090.84	113.6 M
Mask-RT-DETR-M	Inference Model/Trained Model	42.7	78.69 / 78.69	nan / nan	66.6 M
Mask-RT-DETR-S	Inference Model/Trained Model	41.0	33.5007	-	51.8 M
Mask-RT-DETR-X	Inference Model/Trained Model	47.5	114.16 / 114.16	1240.92 / 1240.92	237.5 M
MaskRCNN-ResNet50-FPN	Inference Model/Trained Model	35.6	118.30 / 118.30	nan / nan	157.5 M	Mask R-CNN is a full-task deep learning model from Facebook AI Research (FAIR) that can perform object classification and localization in a single model, combined with image-level masks to complete segmentation tasks.
MaskRCNN-ResNet50-vd-FPN	Inference Model/Trained Model	36.4	118.34 / 118.34	nan / nan	157.5 M
MaskRCNN-ResNet50	Inference Model/Trained Model	32.8	228.83 / 228.83	nan / nan	128.7 M
MaskRCNN-ResNet101-FPN	Inference Model/Trained Model	36.6	148.14 / 148.14	nan / nan	225.4 M
MaskRCNN-ResNet101-vd-FPN	Inference Model/Trained Model	38.1	151.12 / 151.12	nan / nan	225.1 M
MaskRCNN-ResNeXt101-vd-FPN	Inference Model/Trained Model	39.5	237.55 / 237.55	nan / nan	370.0 M
PP-YOLOE_seg-S	Inference Model/Trained Model	32.5	-	-	31.5 M	PP-YOLOE_seg is an instance segmentation model based on PP-YOLOE. This model inherits PP-YOLOE's backbone and head, significantly enhancing instance segmentation performance and inference speed through the design of a PP-YOLOE instance segmentation head.
SOLOv2	Inference Model/Trained Model	35.5	-	-	179.1 M	SOLOv2 is a real-time instance segmentation algorithm that segments objects by location. This model is an improved version of SOLO, achieving a good balance between accuracy and speed through the introduction of mask learning and mask NMS.

**Test Environment Description**: - **Performance Test Environment** - **Test Dataset**: COCO2017 validation set. - **Hardware Configuration**: - GPU: NVIDIA Tesla T4 - CPU: Intel Xeon Gold 6271C @ 2.60GHz - Other Environments: Ubuntu 20.04 / cuDNN 8.6 / TensorRT 8.5.2.2 - **Inference Mode Description** | Mode | GPU Configuration | CPU Configuration | Acceleration Technology Combination | |-------------|----------------------------------------|-------------------|---------------------------------------------------| | Normal Mode | FP32 Precision / No TRT Acceleration | FP32 Precision / 8 Threads | PaddleInference | | High-Performance Mode | Optimal combination of pre-selected precision types and acceleration strategies | FP32 Precision / 8 Threads | Pre-selected optimal backend (Paddle/OpenVINO/TRT, etc.) |

III. Quick Integration¶

❗ Before quick integration, please install the PaddleX wheel package. For detailed instructions, refer to the PaddleX Local Installation Tutorial

After installing the wheel package, a few lines of code can complete the inference of the instance segmentation module. You can switch models under this module freely, and you can also integrate the model inference of the instance segmentation module into your project. Before running the following code, please download the demo image to your local machine.

from paddlex import create_model
model = create_model("PP-YOLOE_seg-S")
output = model.predict("general_instance_segmentation_004.png", batch_size=1)
for res in output:
    res.print()
    res.save_to_img("./output/")
    res.save_to_json("./output/res.json")

After running, the result obtained is:

{'res': "{'input_path': 'general_instance_segmentation_004.png', 'page_index': None, 'boxes': [{'cls_id': 0, 'label': 'person', 'score': 0.8723232746124268, 'coordinate': [88.34339, 109.87673, 401.85236, 575.59576]}, {'cls_id': 0, 'label': 'person', 'score': 0.8711188435554504, 'coordinate': [325.114, 1.1152496, 644.10266, 575.359]}, {'cls_id': 0, 'label': 'person', 'score': 0.842758297920227, 'coordinate': [514.18964, 21.760618, 768, 576]}, {'cls_id': 0, 'label': 'person', 'score': 0.8332827091217041, 'coordinate': [0.105075076, 0, 189.23515, 575.9612]}], 'masks': '...'}"}

The meanings of the running results parameters are as follows: - input_path: Indicates the path of the input image to be predicted. - page_index: If the input is a PDF file, it represents the current page number of the PDF; otherwise, it is None. - boxes: Information of each detected object. - cls_id: Class ID. - label: Class name. - score: Prediction score. - coordinate: Coordinates of the bounding box, in the format [xmin, ymin, xmax, ymax]. - pred: The actual mask predicted by the instance segmentation model. Since the data is too large to be printed directly, it is replaced with ... here. You can use res.save_to_img() to save the prediction results as an image and res.save_to_json() to save the prediction results as a JSON file.

The visualization image is as follows:

Note: Due to network issues, the above URL may not be accessible. If you need to access this link, please check the validity of the URL and try again. If the problem persists, it may be related to the link itself or the network connection.

Related methods, parameters, and explanations are as follows:

create_model instantiates a general instance segmentation model (here, PP-YOLOE_seg-S is used as an example), and the specific explanations are as follows:

Parameter	Parameter Description	Parameter Type	Options	Default Value
`model_name`	Name of the model	`str`	None	`None`
`model_dir`	Path to store the model	`str`	None	None
`threshold`	Threshold for filtering low-confidence objects	`float/None`	None	None

The model_name must be specified. After specifying model_name, the default model parameters built into PaddleX are used. If model_dir is specified, the user-defined model is used.
threshold is the threshold for filtering low-confidence objects. The default is None, which means using the settings from the lower priority. The priority of parameter settings from highest to lowest is: predict parameter > create_model initialization > yaml configuration file.
The predict() method of the general instance segmentation model is called for inference prediction. The predict() method has parameters input, batch_size, and threshold, which are explained as follows:

Parameter	Parameter Description	Parameter Type	Options	Default Value
`input`	Data to be predicted, supporting multiple input types	`Python Var`/`str`/`list`	Python variable, such as image data represented by `numpy.ndarray` File path, such as the local path of an image file: `/root/data/img.jpg` URL link, such as the network URL of an image file: Example Local directory, the directory should contain data files to be predicted, such as the local path: `/root/data/` List, elements of the list must be of the above types of data, such as `[numpy.ndarray, numpy.ndarray]`, `["/root/data/img1.jpg", "/root/data/img2.jpg"]`, `["/root/data1", "/root/data2"]`	None
`batch_size`	Batch size	`int`	Any integer	1
`threshold`	Threshold for filtering low-confidence objects	`float`/`None`	None, indicating the use of settings from the lower priority. The priority of parameter settings from highest to lowest is: `predict parameter > create_model initialization > yaml configuration file` float, such as 0.5, indicating the use of `0.5` as the threshold for filtering low-confidence objects during inference	None

The prediction results are processed, and the prediction result for each sample is of type dict. It supports operations such as printing, saving as an image, and saving as a json file:

Method	Method Description	Parameter	Parameter Type	Parameter Description	Default Value
`print()`	Print the results to the terminal	`format_json`	`bool`	Whether to format the output content using `JSON` indentation	`True`
		`indent`	`int`	Specify the indentation level to beautify the output `JSON` data, making it more readable, only effective when `format_json` is `True`	4
		`ensure_ascii`	`bool`	Control whether to escape non-`ASCII` characters to `Unicode`. If set to `True`, all non-`ASCII` characters will be escaped; `False` retains the original characters, only effective when `format_json` is `True`	`False`
`save_to_json()`	Save the results as a JSON file	`save_path`	`str`	The path to save the file. If it is a directory, the saved file name will be consistent with the input file name	None
		`indent`	`int`	Specify the indentation level to beautify the output `JSON` data, making it more readable, only effective when `format_json` is `True`	4
		`ensure_ascii`	`bool`	Control whether to escape non-`ASCII` characters to `Unicode`. If set to `True`, all non-`ASCII` characters will be escaped; `False` retains the original characters, only effective when `format_json` is `True`	`False`
`save_to_img()`	Save the results as an image file	`save_path`	`str`	The path to save the file. If it is a directory, the saved file name will be consistent with the input file name	None

Additionally, it supports obtaining the visualization image with results and the prediction results through attributes, as follows:

Attribute	Attribute Description
`json`	Get the prediction result in `json` format
`img`	Get the visualization image in `dict` format

For more information on using PaddleX's single-model inference APIs, please refer to the PaddleX Single-Model Python Script Usage Instructions.

IV. Custom Development¶

If you are seeking higher accuracy from existing models, you can use PaddleX's custom development capabilities to develop better instance segmentation models. Before using PaddleX to develop instance segmentation models, please ensure that you have installed the relevant model training plugins for segmentation in PaddleX. The installation process can be found in the custom development section of the PaddleX Local Installation Guide.

4.1 Data Preparation¶

Before model training, it is necessary to prepare the corresponding dataset for each task module. PaddleX provides data verification functionality for each module, and only data that passes the verification can be used for model training. Additionally, PaddleX provides demo datasets for each module, allowing you to complete subsequent development based on the officially provided demo data. If you wish to use a private dataset for subsequent model training, you can refer to the PaddleX Instance Segmentation Task Module Data Annotation Tutorial.

4.1.1 Download Demo Data¶

You can download the demo dataset to a specified folder using the following command:

wget https://paddle-model-ecology.bj.bcebos.com/paddlex/data/instance_seg_coco_examples.tar -P ./dataset
tar -xf ./dataset/instance_seg_coco_examples.tar -C ./dataset/

4.1.2 Data Verification¶

Data verification can be completed with a single command:

python main.py -c paddlex/configs/modules/instance_segmentation/Mask-RT-DETR-L.yaml \
    -o Global.mode=check_dataset \
    -o Global.dataset_dir=./dataset/instance_seg_coco_examples

After executing the above command, PaddleX will validate the dataset and summarize its basic information. If the command runs successfully, it will print Check dataset passed ! in the log. The validation results file is saved in ./output/check_dataset_result.json, and related outputs are saved in the ./output/check_dataset directory in the current directory, including visual examples of sample images and sample distribution histograms.

👉 Details of Validation Results (Click to Expand)

The specific content of the validation result file is:

{
  "done_flag": true,
  "check_pass": true,
  "attributes": {
    "num_classes": 2,
    "train_samples": 79,
    "train_sample_paths": [
      "check_dataset/demo_img/pexels-photo-634007.jpeg",
      "check_dataset/demo_img/pexels-photo-59576.png"
    ],
    "val_samples": 19,
    "val_sample_paths": [
      "check_dataset/demo_img/peasant-farmer-farmer-romania-botiza-47862.jpeg",
      "check_dataset/demo_img/pexels-photo-715546.png"
    ]
  },
  "analysis": {
    "histogram": "check_dataset/histogram.png"
  },
  "dataset_path": "instance_seg_coco_examples",
  "show_type": "image",
  "dataset_type": "COCOInstSegDataset"
}

In the above verification results, check_pass being True indicates that the dataset format meets the requirements. Explanations for other indicators are as follows:

attributes.num_classes: The number of classes in this dataset is 2;
attributes.train_samples: The number of training samples in this dataset is 79;
attributes.val_samples: The number of validation samples in this dataset is 19;
attributes.train_sample_paths: A list of relative paths to the visualized training samples in this dataset;
attributes.val_sample_paths: A list of relative paths to the visualized validation samples in this dataset; Additionally, the dataset verification also analyzes the distribution of sample numbers across all categories in the dataset and generates a distribution histogram (histogram.png):

4.1.3 Dataset Format Conversion/Dataset Splitting (Optional)¶

After completing data verification, you can convert the dataset format or re-split the training/validation ratio by modifying the configuration file or appending hyperparameters.

👉 Details of Format Conversion/Dataset Splitting (Click to Expand)

(1) Dataset Format Conversion

The instance segmentation task supports converting LabelMe format to COCO format. The parameters for dataset format conversion can be set by modifying the fields under CheckDataset in the configuration file. Below are some example explanations for some of the parameters in the configuration file:

CheckDataset:
convert:
enable: Whether to perform dataset format conversion. Set to True to enable dataset format conversion, default is False;
src_dataset_type: If dataset format conversion is performed, the source dataset format needs to be set. The available source format is LabelMe; For example, if you want to convert a LabelMe dataset to COCO format, you need to modify the configuration file as follows:

cd /path/to/paddlex
wget https://paddle-model-ecology.bj.bcebos.com/paddlex/data/instance_seg_labelme_examples.tar -P ./dataset
tar -xf ./dataset/instance_seg_labelme_examples.tar -C ./dataset/

......
CheckDataset:
  ......
  convert:
    enable: True
    src_dataset_type: LabelMe
  ......

Then execute the command:

python main.py -c paddlex/configs/modules/instance_segmentation/Mask-RT-DETR-L.yaml\
    -o Global.mode=check_dataset \
    -o Global.dataset_dir=./dataset/instance_seg_labelme_examples

After the data conversion is executed, the original annotation files will be renamed to xxx.bak in the original path.

The above parameters also support being set by appending command line arguments:

python main.py -c paddlex/configs/modules/instance_segmentation/Mask-RT-DETR-L.yaml\
    -o Global.mode=check_dataset \
    -o Global.dataset_dir=./dataset/instance_seg_labelme_examples \
    -o CheckDataset.convert.enable=True \
    -o CheckDataset.convert.src_dataset_type=LabelMe

(2) Dataset Splitting

The parameters for dataset splitting can be set by modifying the fields under CheckDataset in the configuration file. Some example explanations for the parameters in the configuration file are as follows:

CheckDataset:
split:
enable: Whether to re-split the dataset. When set to True, the dataset will be re-split. The default is False;
train_percent: If the dataset is to be re-split, the percentage of the training set needs to be set. The type is any integer between 0-100, and the sum with val_percent must be 100;

For example, if you want to re-split the dataset with a 90% training set and a 10% validation set, you need to modify the configuration file as follows:

......
CheckDataset:
  ......
  split:
    enable: True
    train_percent: 90
    val_percent: 10
  ......

Then execute the command:

python main.py -c paddlex/configs/modules/instance_segmentation/Mask-RT-DETR-L.yaml \
    -o Global.mode=check_dataset \
    -o Global.dataset_dir=./dataset/instance_seg_labelme_examples

After data splitting, the original annotation files will be renamed as xxx.bak in the original path.

The above parameters can also be set by appending command line arguments:

python main.py -c paddlex/configs/modules/instance_segmentation/Mask-RT-DETR-L.yaml \
    -o Global.mode=check_dataset \
    -o Global.dataset_dir=./dataset/instance_seg_labelme_examples \
    -o CheckDataset.split.enable=True \
    -o CheckDataset.split.train_percent=90 \
    -o CheckDataset.split.val_percent=10

4.2 Model Training¶

A single command can complete model training. Taking the training of the instance segmentation model Mask-RT-DETR-L as an example:

python main.py -c paddlex/configs/modules/instance_segmentation/Mask-RT-DETR-L.yaml \
    -o Global.mode=train \
    -o Global.dataset_dir=./dataset/instance_seg_coco_examples

The following steps are required:

Specify the path to the .yaml configuration file of the model (here it is Mask-RT-DETR-L.yaml,When training other models, you need to specify the corresponding configuration files. The relationship between the model and configuration files can be found in the PaddleX Model List (CPU/GPU))
Specify the mode as model training: -o Global.mode=train
Specify the path to the training dataset: -o Global.dataset_dir. Other related parameters can be set by modifying the fields under Global and Train in the .yaml configuration file, or adjusted by appending parameters in the command line. For example, to specify the first 2 GPUs for training: -o Global.device=gpu:0,1; to set the number of training epochs to 10: -o Train.epochs_iters=10. For more modifiable parameters and their detailed explanations, refer to the PaddleX Common Configuration File Parameters Instructions.

👉 More Details (Click to Expand)

During model training, PaddleX automatically saves the model weight files, with the default being output. If you need to specify a save path, you can set it through the -o Global.output field in the configuration file.
PaddleX shields you from the concepts of dynamic graph weights and static graph weights. During model training, both dynamic and static graph weights are produced, and static graph weights are selected by default for model inference.
After completing the model training, all outputs are saved in the specified output directory (default is ./output/), typically including:
train_result.json: Training result record file, recording whether the training task was completed normally, as well as the output weight metrics, related file paths, etc.;
train.log: Training log file, recording changes in model metrics and loss during training;
config.yaml: Training configuration file, recording the hyperparameter configuration for this training session;
.pdparams, .pdema, .pdopt.pdstate, .pdiparams, .pdmodel: Model weight-related files, including network parameters, optimizer, EMA, static graph network parameters, static graph network structure, etc.;

4.3 Model Evaluation¶

After completing model training, you can evaluate the specified model weights file on the validation set to verify the model's accuracy. Using PaddleX for model evaluation can be done with a single command:

python main.py -c paddlex/configs/modules/instance_segmentation/Mask-RT-DETR-L.yaml \
    -o Global.mode=evaluate \
    -o Global.dataset_dir=./dataset/instance_seg_coco_examples

Similar to model training, the following steps are required:

Specify the .yaml configuration file path for the model (here it is Mask-RT-DETR-L.yaml)
Specify the mode as model evaluation: -o Global.mode=evaluate
Specify the path to the validation dataset: -o Global.dataset_dir. Other related parameters can be set by modifying the Global and Evaluate fields in the .yaml configuration file. For details, refer to PaddleX Common Model Configuration File Parameter Description.

👉 More Details (Click to Expand)

When evaluating the model, you need to specify the model weights file path. Each configuration file has a default weight save path built-in. If you need to change it, simply set it by appending a command line parameter, such as -o Evaluate.weight_path=./output/best_model/best_model.pdparams.

After completing the model evaluation, an evaluate_result.json file will be generated, which records the evaluation results, specifically whether the evaluation task was completed successfully and the model's evaluation metrics, including AP.

4.4 Model Inference and Model Integration¶

After completing model training and evaluation, you can use the trained model weights for inference prediction or Python integration.

4.4.1 Model Inference¶

To perform inference prediction via the command line, simply use the following command. Before running the following code, please download the demo image to your local machine.

python main.py -c paddlex/configs/modules/instance_segmentation/Mask-RT-DETR-L.yaml \
    -o Global.mode=predict \
    -o Predict.model_dir="./output/best_model/inference" \
    -o Predict.input="general_instance_segmentation_004.png"

Similar to model training and evaluation, the following steps are required:

Specify the .yaml configuration file path of the model (here it's Mask-RT-DETR-L.yaml)
Specify the mode as model inference prediction: -o Global.mode=predict
Specify the model weights path: -o Predict.model_dir="./output/best_model/inference"
Specify the input data path: -o Predict.input="...". Other related parameters can be set by modifying the Global and Predict fields in the .yaml configuration file. For details, refer to PaddleX Common Model Configuration File Parameter Description.

4.4.2 Model Integration¶

The model can be directly integrated into the PaddleX Pipeline or into your own project.

1.Pipeline Integration

The instance segmentation module can be integrated into the General Instance Segmentation Pipeline of PaddleX. Simply replace the model path to update the instance segmentation module of the relevant pipeline.

2.Module Integration The weights you produce can be directly integrated into the instance segmentation module. Refer to the Python example code in Quick Integration , and simply replace the model with the path to your trained model.

You can also use the PaddleX high-performance inference plugin to optimize the inference process of your model and further improve efficiency. For detailed procedures, please refer to the PaddleX High-Performance Inference Guide.