s6.app.cog.keypoint¶
Keypoint training, preview, ONNX export, and TensorRT deployment utilities.
- s6.app.cog.keypoint.load_dataset_config(config_path: str) Config
- s6.app.cog.keypoint.effective_num_segmentation_classes(config: Config) int | None
Return segmentation classes only when segmentation targets are enabled.
- s6.app.cog.keypoint.build_model(config: Config, pretrained_backbone: bool = False) GenericKeypointModel
- s6.app.cog.keypoint.unpack_batch(batch) Tuple[Tensor, Tensor, Tensor | None, Tensor | None]
- s6.app.cog.keypoint.unpack_model_outputs(outputs) Tuple[Tensor, Tensor, Tensor | None]
- s6.app.cog.keypoint.log_graph(writer, model, dataloader, device)
- s6.app.cog.keypoint.should_log_verbose(args: Namespace, writer, iteration: int) bool
Return whether histogram and image logging should run for an iteration.
- s6.app.cog.keypoint.load_training_cameras(config: Config) Dict[str, Camera]
Load cameras required by configured stereo training losses.
- s6.app.cog.keypoint.triangulation_reprojection_loss(predicted_keypoints: Tensor, transforms: Tensor, cameras: Dict[str, Camera], left_key: str, right_key: str, output_size: int) Tensor | None
Return stereo reprojection consistency loss for even/odd LL/LR pairs.
- s6.app.cog.keypoint.triangulation_rigidity_loss(predicted_keypoints: Tensor, transforms: Tensor, cameras: Dict[str, Camera], left_key: str, right_key: str, output_size: int) Tensor | None
Return rigid segment-length loss for reconstructed T1 triplets.
- s6.app.cog.keypoint.batch_draw_keypoints(images: Tensor, keypoints: Tensor, masks: Tensor | None = None) Tensor
- s6.app.cog.keypoint.train_step(batch, model, optimizer, keypoint_criterion, segmentation_criterion, device, segmentation_loss_weight: float = 1.0, precision: str = 'fp32', loss_terms: LossTermsConfig | None = None, cameras: Dict[str, Camera] | None = None, stereo_pairing=None, output_size: int | None = None)
- s6.app.cog.keypoint.make_train_dataloader(args)
- s6.app.cog.keypoint.load_deploy_output_size(config_path: str) int
- s6.app.cog.keypoint.resolve_precision(args: Namespace) str
- s6.app.cog.keypoint.resolve_deploy_path(args: Namespace, config: Config) str
- s6.app.cog.keypoint.resolve_trt_engine_path(onnx_path: str) str
Return the TensorRT engine path derived from an ONNX export path.
- s6.app.cog.keypoint.run_trtexec_conversion(args: Namespace, onnx_path: str) str
Convert an exported ONNX model into a TensorRT engine with trtexec.
- s6.app.cog.keypoint.train(args)
- s6.app.cog.keypoint.preview_dataset(args)
- s6.app.cog.keypoint.deploy_model(args)
- s6.app.cog.keypoint.main()