What's new in version 1.7.1

New

Samples

• Automate Road Surface Investigation Using Deep Learning
• Detecting and Categorizing Brick Kilns from Satellite Imagery
• Extracting Slums from Satellite Imagery
• Extracting Sinkholes from Aerial Imagery
• Detecting Settlements Using Supervised Classification and Deep Learning
• How much green is Delhi as on 15 Oct 2017?
• Maximizing Fire Protection Coverage
• Calculate Impervious Surfaces from Multispectral Imagery using Deep Learning
• Safe Streets to Schools
• Spatial and temporal distribution of service calls using big data tools
• Time Series Prediction of AirBnB Properties in New York City

Module Changes

arcgis.raster.functions.gbl

New functions

• corridor

arcgis.env

• new parameter type_init_tail_parameters (See arcgis.learn below.)

arcgis.learn

• Adds support for multiple GPU machines to learn module
• Adds the DeepLabV3 model based on torchvision
• includes multispectral data support
• Enhancements to Multispectral data support:
• turn on DRA with the statistics_type parameter in various show_results() and show_batch() functions
• adds environment variable type_init_tail_parameters to control Model Tail initialization for arcgis.learn functions
• Sanctions classify_objects() over FeatureClassifier.categorize_features() for feature categorization
• Adds support for training models on either cpu or gpu devices
• Adds support for evaluating MaskRCNN model performance with correct metrics for the trained model to compare the results
• Adds accuracy() function to UnetClassifier
• Adds unet_aux_loss parameter to the PSPNetClassifier
• Adds support for training a subset of classes from prepare_data() class_mapping parameter to MaskRCNN
• Adds multispectral data support to:
• FeatureClassifier
• SingleShotDetector
• RetinaNet
• PSPNetClassifier
• MaskRCNN
• Adds support for resnet18 and resnet34 backbones to MaskRCNN
• Adds support for batchnorm unfreezing in * PSPNetClassifier
• Adds support for panchromatic data
• Adds validation to class_mapping parameter for prepare_data() function

Fixes

Module Changes

arcgis.gis.admin

• Fixes formatting in EnterpriseUsers.update() function

arcgis.mapping

• Fixes error when drawing map widget using Microsoft Edge

arcgis.raster.functios.gbl

• General improvements to documentation
• Improves alphabetical ordering of API Reference

arcgis.learn

• Improves error messaging when calling from_model() if proper libraries are not installed
• Deprecates the FeatureClassifier.categorize_features() method in favor of classify_objects()
• Improves messaging when input_video_path parameter video does not exist for RetinaNet.predict_video() and SingleShotDetector.predict_video() functions
• Fixes bug when re-running a previously saved SingleShotDetector model
• Fixes various model inferencing errors when using Image Server
• Fixes errors when using detect_objects() when raster function configuration information is inaccessible
• Fixes error when running detect_objects() when using no model_arguments
• Fixes EntityRecognizer.extract_entities() returning a data frame with an empty column name
• Improves messaging when incorrect path passed as path argument to prepare_data()
• Fixes error when list of tensors is empty when running SingleShotDetector.fit() model
• Fixes model accuracy function in Unet and PSPNet to return maximum accuracy if checkpoint is True
• Improves tagging scheme documentation for prepare_data() function
• Improves visual accuracy when using Multispectral imagery with UNetClassifier
• Fixes missing supported_backbone documentation for all models
• Improves predict() and from_model() documentation on the FeatureClassifier
• Fixes [show_batch()] errors on data objects when class_mapping parameter values from prepare_data()are non-contiguous
• Fixes errors with show_results() and fit() methods of EntityRecognizer
• Fixes error when prepare_data() dataset_type argument value is BILUO
• Fixes extract_entities() error when reading UTF-16 encoded files
• Fixes to load() function for input paths
• Fixes error reading file names in EntityRecognizer.load()
• Fixes error causing accuracy() to always return 1 with certain class_mapping values when using UnetClassifier
• Fixes error where color map values were truncated using MaskRCNN
• Fixes error in SingleShotDetector.save() method by adding optional overwrite parameter