Changes

  • 11.3.0
    • Added the super resolution method,realizing the process of model inference based on Real-Esrgan models.
    • Added model evaluation support for image super resolution, which can be used to evaluate model accuracy
    • Fixed the problem of the model’s video memory not being released after the completion of prompt segmentation inference.
  • 11.2.1
    • Fixed the problem of object extraction continue training.
  • 11.2.0
    • Added the prompt segmentation tool based on SAM model.
    • ReDet model is added to the training function of object detection.
    • Mask2Former model is added to the training function of multi-classification.
    • Object detection and object extraction add the ability to support multi-band images in the whole process.
  • 11.1.0
    • Segformer model is added to the training function of binary classification and multi classification.
    • Siam-Segformer model is added to the training function of general change detection.
    • RTMDet model is added to the training function of object detection.
    • RTMDet-Ins model is added to the training function of object extraction.
    • Support batch generation of training data.
    • Model training adds the ability to continue training.
    • Model training adds the ability to transfer learning.
    • New learning rate adaptive adjustment strategy for model training
    • Added support for batch data inference.
    • Support inference by scope.
    • Binary classification, multi object classification, general change detection support the full process of using 8-bit multi-band image to generate training data, model training, and model inference.
    • Added model conversion capability
  • 11.0.0
    • Added the general_change_detection method, realizing the whole process of training data creation, model training and model inference based on Siam-SFNet and DSAMNet models
    • Added model_evaluation method for object detection, binary classification, multi-classification, general change detection, which can be used to evaluate model accuracy
    • Upgrade Conda environment to support Nvidia RTX30 series graphics cards to meet product hardware adaptation requirements
    • New SFNet model for binary classification and multi-classification to enhance the ability to capture contextual and semantic spatial information
    • Added learning rate decay strategy, which can be used to improve model training effect
    • Interval_validation is added to the training_config file to control the validation interval during model training, and reduce the total training time
    • Added the init_data parameter in model training to control the division of training sets and verification sets, facilitating comparison of model effects
    • Supports the training of multiple GPU devices and also support specified GPU device number, and supports more efficient GPU parallel computing, which can be used to improve model training accuracy and training efficiency
    • Provide a variety of pre-training models such as binary classification of water bodies and buildings and land cover multi_classification, so as to reduce the use cost of users
  • 10.2.1
  • 10.1.1
    • Added training function for binary classification D-LinkNet model
    • Added the function of generating graph spatiotemporal regression training data
    • Fix the problem of continuous occupation of video memory after ground object classification model inference
    • Improve the support for Chinese in the object extraction training data path
    • Improve the support for pre-training models in the object extraction training process
    • Improve the support for the plane coordinate system of the object extraction results
  • 10.1.0
    • Added two-dimensional network analysis and three-dimensional network analysis modules
    • Add address matching module
    • Added a variety of grid comprehensive functions: including mode filtering, expansion, contraction, cannibalization, region grouping, boundary cleaning, refinement, etc.
    • Added geographic simulation functions, including cellular automata based on principal component analysis (PCA) and artificial neural network (ANN)
    • Improve the data import function, and add support for importing file catalog functions
    • Added thematic map function
    • Added image data object extraction function, involving training data generation, model training, and model inference interfaces
    • Added FPN, DeepLab V3+ model support for binary classification of image data
    • Added image classification function for image data, involving model training and model inference
    • Added a new model conversion interface to support the conversion of image classification training result models to mobile models
    • Added image data target detection function, involving model training and model inference interface
    • Added traffic flow prediction function based on DCRNN
    • Added structured data classification function based on gradient boosting
  • 10.0.1