Features: LibTorch integration for UE5: runtime links LibTorch, stages DLLs, delay-loads CUDA, runs TorchScript in packaged builds.

Inference examples: ResNet-18 classification and YOLOv5s detection with labels, NMS, IoU tracking; Blueprint-callable (legacy combined actor + new minimal classification/tracking actors).

Data bridge utilities: render target capture/image loading to CPU, NCHW tensor prep, YOLOv5 decoding.

Code Modules: LibtorchIntegration (Runtime)

Blueprints: 2 (BP_LibtorchInferenceActor, WB_Overlay)

C++ Classes: 6 (ALibtorchExampleActor, ALibtorchInferenceActor, ALibtorchClassificationActor, ALibtorchTrackingActor, ALibtorchMapTriggerActor, FSimpleIoUTracker)

Helper Python scripts: check_cuda.py, export_resnet18_ts.py, export_yolov5s_ts.py

Network Replicated: No

Required Plugins:

Enhanced Input
Modeling Tools Editor Mode

Dependencies (included): libtorch-win-shared-with-deps-2.9.1+cu128

Supported Platforms: Windows (x64)

Documentation:

Path (relative to plugin root): README.md
Covers setup, actor usage (minimal/legacy/example), demo maps, pipeline, troubleshooting.

Notes: Ships full runtime wiring so users can drop in their own .pt models. Includes sample maps/assets and ResNet-18/YOLOv5s references.

Discord: https://discord.gg/bKn8fzG8

Altered Horizon Ventures Inc

Altered Horizon Ventures Inc

产品类型