Get expected MD5 hash for a model by filename. Args: filename: The model filename Returns: MD5 hash string i

Retrieve a model context from the provided architecture configuration. Args: config: Architecture configuration. Returns

Return a fresh ``_DummyModel`` instance.

Retrieve the configuration parameters used by the model.

Return the current keypoint schema inferred from the active-keypoint mask.

Retrieve the configuration parameters that will be used for training.

Sample points in [0, 1] x [0, 1] coordinate space based on their uncertainty. The uncertainties are calculated for each point using 'uncertainty_

This method allows to interpolate the pre-trained position encodings, to be able to use the model on higher resolution images. This implementa

Yield the configured content chunks.

List all available model filenames. Returns: List of model filenames

Load engine.

Load onnx from file.

Download, validate, instantiate all models, and test from_checkpoint round-trip.

Download pre-trained weights if they are not already downloaded. Bare filenames (no directory component, e.g. ``rf-detr-base.pth``) are resol

Apply AdaLN modulation to a feature tensor. Args: features: Input feature map to modulate. scale: Per-feature scale terms.

Accumulate predictions and matching data for one test batch. Mirrors :meth:`on_validation_batch_end` for the test evaluation loop triggered b

Update EMA after optimizer steps.

Plot aggregate and component training losses from a PTL ``metrics.csv`` file. Reads the CSV written by PyTorch Lightning's ``CSVLogger``, groups

Format bounding-box predictions as COCO result dicts.

Format keypoint predictions as COCO result dicts.

Format segmentation mask predictions as COCO result dicts.

Find the largest per-device batch size that fits in memory for one train step. Uses exponential search (1, 2, 4, ...) up to the first failure, th

Raise the configured HTTP error, if any.

Recommend gradient accumulation steps to reach target effective batch size. Args: safe_micro_batch: Per-step batch size that fits in memo

Reset keypoint Gaussian precision outputs to unit values. Keypoint channels 4, 5, and 6 encode the lower-triangular precision Cholesk

Resolve batch_size='auto' into concrete batch_size and grad_accum_steps using a probe. Expects model_context to have attributes: .device (torch.d

Reset random, numpy, torch, and CUDA seeds before each test. Defaults to seed 7. Override per-test via indirect parametrize:: @pytest.ma

Initialise the averaged model at fit start. Args: trainer: The Lightning Trainer instance. pl_module: The ``RFDETRMod

Loss used in RetinaNet for dense detection: https://arxiv.org/abs/1708.02002. Args: inputs: A float tensor of arbitrary shape.

Materialize the current sample as a supervision ``Detections`` object.

Convert to dictionary, filtering out zero ratios.

Update drop_path rates for backbone encoder layers with linear schedule. Applies a linear schedule where the first layer has drop_path_rate=0

Validate MD5 hash of pretrained weights file. Args: pretrain_weights: Path to the pretrained weights file strict: If True, raise

Normalize image and convert target coordinates to relative format. Applies ImageNet-style channel normalization to the image, then converts

Apply the Albumentations transform to image and target. This method handles the data format conversion between RF-DETR and Albumentations:

Convert image and YOLO detections to RF-DETR format. Args: image: PIL Image target: dict with 'image_id' and 'detecti

Export *model* to *output_path*. Args: model: The PyTorch model to export. input_shape: Input tensor shape (excluding

Include lazy re-exports in interactive discovery.

Return self to satisfy context manager protocol.

Propagate exceptions raised inside the context.

Lazily resolve training/PTL and plus-only exports and handle removed-module aliases. This hook is only invoked on explicit attribute access (e.g.

Lazily resolve legacy re-exports without creating import-order cycles.

Lazy failure for missing plus exports: warn on import, raise on access.

Return the item at the (possibly remapped) index.