Detailed Description of Changes in HALCON 20.05.0.1 Progress (Hotfix)
The changes in HALCON 20.05.0.1 Progress are described with respect to HALCON 20.05.0.0 Progress.
Deep Learning
- If the system parameter "cudnn_deterministic" was specified as "true" and the model parameter "optimize_for_inference" was also "true", the training of this deep learning model did not use the corresponding deterministic cuDNN backward algorithm for convolution layers as required. Hence, the training of such a model was most likely not deterministic on a GPU. This problem has been fixed.
Matching
- find_shape_model, find_shape_models, find_scaled_shape_model, find_scaled_shape_models, find_aniso_shape_model, find_aniso_shape_models, and find_shape_model_3d did not abort in a timeout error in rare cases where the operator was parallelized and a timeout was set. This problem has been fixed.
Licensing
- When using a license bound to a dongle, HDevelop could periodically fail to communicate with the dongle and complain about a licensing error in response. This problem has been fixed.
Release Notes of Previous HALCON Versions
Follow this link to read about the changes of previous HALCON versions.