| license: mit | |
| library_name: pytorch | |
| pipeline_tag: image-segmentation | |
| tags: | |
| - lane-detection | |
| - carla | |
| - tensorrt | |
| - onnx | |
| # RCLane release model (17-09-2026) | |
| Model artifacts used with the RCLane source release at Git commit | |
| [`833455bf0046986531913ae7abcbc0dc1fd3e38f`](https://github.com/shinlnh/RCLane/commit/833455bf0046986531913ae7abcbc0dc1fd3e38f). | |
| ## Files | |
| | File | Purpose | SHA-256 | | |
| | --- | --- | --- | | |
| | `rclane_b0_e19_fp16.engine` | Exact FP16 TensorRT engine used by the deployed lane-detection service | `1dd1ea149559dd5694e450067cd1d34a2299c49973eb91fb06a07bce78037e72` | | |
| | `rclane_b0_e19_trt85.onnx` | Portable ONNX graph prepared for TensorRT 8.5 | `8847f7da77c5bcdb67c166e119f9817f68aa2d420acbd333128a10826ce363cd` | | |
| | `rclane_b0_e19.pth` | PyTorch training checkpoint, CARLA B0 epoch 19 | `ef3eb57a7064cb89783144a05344635e1be3a7527dd69a2022b0240782908866` | | |
| | `manifest.json` | Artifact metadata and checksums | — | | |
| ## Interface | |
| - Input: `images`, shape `N x 3 x 320 x 800` | |
| - Outputs: `seg_map`, `up_arrow`, `down_arrow`, `up_bound`, `down_bound` | |
| - ONNX opset: 17 | |
| TensorRT engines are environment-specific. The included engine is the exact deployed | |
| artifact and was used on an NVIDIA GeForce RTX 3050 6GB Laptop GPU (compute capability | |
| 8.6). Rebuild the engine from the ONNX file when using a different GPU, TensorRT, CUDA, | |
| or driver stack. | |