1 vin.backbones.evl

vin.backbones.evl

EVL backbone adapter for VIN scorer architectures.

VIN consumes raw EFM snippet dicts as input and queries EVL’s 3D neck features to score candidate poses. This module owns the external EVL dependency boundary; model definitions depend on EvlBackboneConfig rather than importing EVL internals directly.

EVL outputs are checkpoint-local actor-visible predictions and features, not ground truth. model_cfg and model_ckpt identify the runtime assets by resolved path, but this adapter does not calculate content hashes or attach them to EvlBackboneOutput; durable cache writers must record that identity.

1.1 Classes

Name Description
EvlBackbone Run EVL and expose actor-visible feature volumes in their voxel frame.
EvlBackboneConfig Select the EVL model assets, device, and exported feature families.

1.1.1 EvlBackbone

vin.backbones.evl.EvlBackbone(config)

Run EVL and expose actor-visible feature volumes in their voxel frame.

freeze controls gradients, but inference always executes under torch.no_grad and puts EVL in evaluation mode. The wrapper retains the resolved asset paths on config; returned tensors carry no independent checkpoint or config identity.

1.1.1.1 Methods

Name Description
forward Run EVL and return the feature volumes needed by VIN.
1.1.1.1.1 forward
vin.backbones.evl.EvlBackbone.forward(efm)

Run EVL and return the feature volumes needed by VIN.

1.1.1.1.1.1 Parameters
Name Type Description Default
efm Mapping[str, Any] Raw EFM snippet dict (unbatched T×... or batched B×T×...). required
1.1.1.1.1.2 Returns
Name Type Description
EvlBackboneOutput Actor-visible EvlBackboneOutput whose voxel tensors are
EvlBackboneOutput Tensor["B C D H W"]-like fields aligned to
EvlBackboneOutput PoseTW["B 12"] world-from-voxel poses. Exact feature semantics
EvlBackboneOutput remain local to this config/checkpoint pair.

1.1.2 EvlBackboneConfig

vin.backbones.evl.EvlBackboneConfig()

Select the EVL model assets, device, and exported feature families.

The resolved model_cfg and model_ckpt paths jointly determine the learned evidence semantics for one run. Paths are not immutable provenance: callers that persist outputs must also capture config and checkpoint hashes.

1.1.2.1 Attributes

Name Description
target_type Factory target for BaseConfig.setup_target.
paths Project path resolver.
model_cfg Resolved Hydra YAML used to instantiate efm3d.model.evl.EVL.
model_ckpt Resolved checkpoint containing an EVL state_dict payload.
device Torch device to run EVL on (auto-selects CUDA if available).
freeze Disable gradients for all EVL parameters when True.
features_mode Which EVL features to expose.
free_input_mode Source of free-space input: native EVL output or explicit derivation.

1.2 Functions

Name Description
filter_backbone_output_for_features_mode Apply EvlBackboneConfig.features_mode to an EVL output container.
resolve_free_input Resolve EVL free-space evidence under the closed source contract.

1.2.1 filter_backbone_output_for_features_mode

vin.backbones.evl.filter_backbone_output_for_features_mode(
    output,
    *,
    features_mode,
)

Apply EvlBackboneConfig.features_mode to an EVL output container.

1.2.1.1 Parameters

Name Type Description Default
output EvlBackboneOutput Full EVL output container before feature-family filtering. required
features_mode Literal['heads', 'neck', 'both'] Feature family requested by the caller. required

1.2.1.2 Returns

Name Type Description
EvlBackboneOutput Filtered output with unsupported feature families set to None or
EvlBackboneOutput empty dictionaries.

1.2.2 resolve_free_input

vin.backbones.evl.resolve_free_input(mode, native_free_input, counts, occ_input)

Resolve EVL free-space evidence under the closed source contract.

Back to top