Mech-DLK SDK 2.0.0 Release Notes
New Features
-
Provided efficient and highly scalable C APIs for inference based on models exported from Mech-DLK. Supported algorithms are as follows:
-
Classification
-
Object detection
-
Instance segmentation
-
Defect segmentation
-
Fast positioning
-
-
Allows for various inference backends and high-performance deployment on devices using GPU or CPU. The available inference backends include the following:
-
GpuDefault: inference using the default GPU mode.
-
GpuOptimization: Inference using the GPU (Optimization) mode. In this mode, it takes one to five minutes for model optimization, and the floating-point precision format can be set according to actual situation.
-
CPU: model deployment based on CPU (for devices without a GPU).
-
-
Enabled inference based on cascaded models to simplify the deployment of complex projects.
-
Enabled simultaneous inference of multiple images to enhance inference efficiency.