Python API catalog

Find a class or function by name, description or module. Filter by category to explore a part of the toolkit. Signatures and parameter documentation are generated from the source tree at build time.

Use Import surface to distinguish package exports (such as helia_edge.metrics.ConfusionMatrix) from APIs defined in individual modules. Public import paths are searchable. Class pages link inherited methods to their defining class.

Backend support varies by component. Read the backend support guide before choosing a TensorFlow or PyTorch workflow.

322 of 322 APIs

Search Python API

Name	Description
AddSineWaveAugmentation	Add a fixed sine wave in training and inference.
AmplitudeWarpAugmentation	Apply amplitude warping to the 1D input.
append_layersArchitectures	Appends layers to a model by cloning it and adding the layers.
att_blockArchitectures	Attention block
attention_token_mixerArchitectures	Token mixer using multi-head attention
AugmentationPipelineAugmentation	Apply layers in order to a tensor or structured sample.
BackendUnavailableConversion	The active Keras backend cannot export: LiteRT export needs the TensorFlow backend.
BaseAugmentationAugmentation	Base for rank-specific transforms, with one sampling/application contract.
BaseAugmentation1DAugmentation	One-dimensional signals with optional batch axis.
BaseAugmentation2DAugmentation	Two-dimensional images with optional batch axis.
BaseAugmentationParamsAugmentation	Shared construction-time configuration; never validates live tensors.
batch_normalizationLayers	Batch normalization
buildArchitectures	Build a Composer model.
buildArchitectures	Build a Conformer model.
buildArchitectures	Build a ConvMixer model.
buildArchitectures	Construct an untrained CorNET regressor with one linear output.
buildArchitectures	Build a EfficientNetV2 model.
buildArchitectures	Construct an untrained one-frame FastEnhancer with named streaming inputs and outputs.
buildArchitectures	Build a MetaFormer model.
buildArchitectures	Construct an untrained MiniResNet-v1 classifier for NHWC spectrogram patches.
buildArchitectures	Build a faithful MLPerf Tiny architecture without loading or converting weights.
buildArchitectures	Build a MobileNetV1 model.
buildArchitectures	Build a MobileOne model.
buildArchitectures	Build a RegNet model.
buildArchitectures	Build a ResNet model.
buildArchitectures	Build the Silero VAD v6 16 kHz streaming model, untrained.
buildArchitectures	Build the model a spec describes.
buildArchitectures	Build a TCN model.
buildArchitectures	Construct an untrained TimePPG regressor with one linear output.
buildArchitectures	Build a TsMixer model.
buildArchitectures	Build a UNet model.
buildArchitectures	Build a UNext model.
CalibrationRecordConversion	The calibration array as stored (``.npy``), its length and, for a streaming model, the state resets.
CascadedBiquadFilterPreprocessing	Implements a 2nd order cascaded biquad filter using direct form 1 structure.
check_calibrationConversion	Reject missing, unexpected or malformed calibration data.
check_named_calibrationConversion	Reject calibration for a model with several inputs unless it names every input once.
check_resetsConversion	Refuse state resets that are not increasing steps from 1 to ``steps - 1`` of a streaming model.
compact_tcn_paramsArchitectures	Four small SE4 blocks with 1/2/4/8 dilations and per-point linear output.
composer_layerArchitectures	Composes a sequential set of networks/layers.
ComposerLayerParamsArchitectures	Composer layer parameters
ComposerParamsArchitectures	Composer Network parameters
compute_checksumUtilities	Compute checksum of file.
compute_metricsMetrics	Compute set of metrics for y_true and y_pred.
conformer_blockArchitectures	Conformer block
conformer_layerArchitectures	Conformer functional layer
ConformerBlockParamsArchitectures	Conformer block parameters
ConformerParamsArchitectures	Conformer parameters
confusion_matrixMetrics	Compute confusion matrix using keras w/ addition to normalize.
confusion_matrix_plotPlotting	Generate confusion matrix plot via matplotlib/seaborn
ConfusionMatrixMetrics	Accumulate class counts and return a row-normalized confusion matrix.
ContrastiveTrainerTraining	Creates a self-supervised contrastive trainer for a model.
conv_blockArchitectures	Convolutional block
conv_mixer_blockArchitectures	ConvMixer block
conv_mixer_layerArchitectures	ConvMixer: https://openreview.net/pdf?id=TVHS5Y4dNvM.
conv_token_mixerArchitectures	Token mixer using separable convolution
conv1dLayers	1D convolutional layer using 2D convolutional layer
conv2dLayers	2D convolutional layer
ConversionModeConversion	How the Keras model is traced for conversion.
convert_inputs_to_tf_datasetData	Convert inputs to tf.data.Dataset.
convert_litertConversion	Convert a Keras model to LiteRT bytes; ``export_model`` validates the arguments first.
ConvMixerParamsArchitectures	ConvMixer parameters
CorNetParamsArchitectures	CorNET heart-rate regressor: two convolution stages then stacked LSTMs.
create_dataset_from_dataData	Helper function to create dataset from static data
create_interleaved_dataset_from_generatorData	Adapt caller-owned schedules to tf.data without changing sample weights.
DataSourceData	Records addressed by index, such as a list or a Grain source.
disable_tensorflow_gpuUtilities	Disable TensorFlow GPU
DistillerTraining	Train a student using target labels and a teacher's softened predictions.
download_fileUtilities	Download file from supplied url to destination streaming.
download_s3_fileUtilities	Download a file from S3
download_s3_objectUtilities	Download an object from S3
download_s3_objectsUtilities	Download all objects in a S3 bucket with a given prefix.
download_s3_prefixUtilities	Download all objects under an S3 prefix into a local directory.
efficientnet_coreArchitectures	EfficientNet core
EfficientNetParamsArchitectures	EfficientNet parameters
efficientnetv2_layerArchitectures	Create EfficientNet V2 TF functional model
EmaResidualVectorQuantizerLayers	Residual VQ with EMA codebook updates.
env_flagUtilities	Return the specified environment variable coerced to a bool, as follows:
environment_differencesConversion	How ``current`` differs from ``recorded``, in the versions that can change exported bytes.
environment_recordConversion	Describe the running environment.
EnvironmentEntryConversion	Versions that can change exported bytes.
EnvironmentRecordConversion	`dataclass`
exportConversion	Export a Keras model to LiteRT with its export record (``helia-edge/export-record@1``).
ExportConversion	`dataclass`
export_createPackage	Build SPEC, load its weights, export it and write model.tflite, model.weights.h5 and record.json.
export_litertConversion	Export with an already validated spec and calibration array.
export_modelConversion	Export a Keras model.
export_reproducePackage	Export a record's model again and compare.
export_schemaPackage	Print the JSON Schema of the export record (helia-edge/export-record@1).
ExportOptionsConversion	Format options of a LiteRT export.
ExportRecordConversion	One exported artifact: the model, its weights, how it was exported, and the result.
ExportResultConversion	`dataclass`
ExportSettingsConversion	How the artifact was exported. ``batch_size`` is the batch of every model input.
ExportSpecConversion	What to export. ``precision``, ``io_dtype`` and ``mode`` have no defaults.
family_mappingConversion	The weight mapping ``name`` of the spec's family (its params module's ``MAPPINGS``).
fastenhancer_mappingArchitectures	Mapping of every weight of ``build(params)`` from folded tensors in ONNX export layout.
fastenhancer_weight_shapesArchitectures	Folded tensors in ONNX/PyTorch layout, keyed by module path.
FastEnhancerCompressionArchitectures	Drop the Nyquist bin and compress magnitude: x * max(|x|, 1e-5)^(c - 1).
FastEnhancerFrequencyAttentionArchitectures	Multi-head self-attention across frequency bands, no output projection.
FastEnhancerFrequencyProjectionArchitectures	Project the frequency axis of (batch, freq, channels) with a [in, out] kernel.
FastEnhancerGRUStepArchitectures	One GRU step per frequency band, gates z, r, h and reset after matmul.
FastEnhancerMaskOutputArchitectures	Apply a complex mask, undo compression and restore a zero Nyquist bin.
FastEnhancerParamsArchitectures	Folded-inference FastEnhancer config for one spectral frame per call.
FastEnhancerPositionalEmbeddingArchitectures	Add a learned (freq, channels) embedding.
FastEnhancerResolvedConfigArchitectures	Serializable record of a preset, its overrides and the concrete config.
FastEnhancerRNNFormerParamsArchitectures	RNNFormer stage: per-band GRU over time, then attention across bands.
fc_blockArchitectures	Fully connected block
file_recordConversion	The record of ``data`` written as ``name``.
file_sha256Conversion	sha256 of a file's bytes.
FileRecordConversion	A file written next to the record, by name, sha256 and size.
FirFilterPreprocessing	Apply FIR filter to the input.
FrequencyMixStyle2DAugmentation	Apply frequency mix style augmentation to the 2D input.
GateReorderPackage	Reorder equal gate blocks along ``axis``, for example ONNX LSTM ``"iofc"`` to Keras ``"ifco"``.
geluLayers	GeLU activation layer
generate_bottleneck_blockArchitectures	Generate functional bottleneck block.
generate_residual_blockArchitectures	Generate functional residual block
get_butter_sosPreprocessing	`cached`
get_flopsMetrics	Calculate FLOPS for keras.Model or keras.Sequential.
get_output_signatureData	Get output signature from sample outputs
get_output_signature_from_fnData	Get output signature from a function
get_output_signature_from_genData	Get output signature from a generator
get_predicted_threshold_indicesMetrics	Get prediction indices that are above threshold (confidence level).
gluLayers	Gated linear unit layer
golden_npzConversion	A golden@2 sequence for a streaming model: ``frames`` are consecutive calls of its signal input.
GoldenRecordConversion	A ``helia-model-zoo/golden@2`` sequence: ``steps`` calls with the state carried and reset at
gradient_stepTraining	Differentiate ``loss_fn`` with the active backend and apply ``model.optimizer`` once.
GSAutoencoderTraining	Convenience wrapper around (encoder -> GumbelSoftmaxBottleneck -> decoder).
GumbelSoftmaxBottleneckLayers	Discrete bottleneck via Gumbel-Softmax (Concrete) with optional straight-through hard one-hot.
hard_sigmoidLayers	Hard sigmoid activation layer
helia_exportUtilities	Register serializable symbols; path is retained for API compatibility.
HeliaEdgeRecordConversion	The helia-edge that exported: its version, how the version identifies the code, and the commit
import_weightsPackage	Set every weight of ``model`` from the file at ``path`` as ``mapping`` describes.
imported_sourceConversion	The record of weights imported through ``mapping`` from a file with ``sha256``.
ImportReportPackage	`dataclass`
infoPackage	Print versions, the KERAS_BACKEND variable and installed optional capabilities.
inspect_modelPackage	Print a .tflite model's inputs, outputs and operators as JSON.
interval_maskAugmentation	Sample one interval per example; returned mask broadcasts over channels.
IODTypeConversion	Element type of an exported model's inputs and outputs.
IORecordConversion	The artifact's inputs and outputs in subgraph order, read back from the artifact.
l2_normalizeLosses	Performs L2 normalization on a tensor along a given axis.
layer_normalizationLayers	Layer normalization
LayerNormalizationLayers	Keras ``LayerNormalization`` that also normalizes non-trailing axes on the Torch backend.
LayerNormalization1DPreprocessing	Apply Layer Normalization to the input.
LayerNormalization2DPreprocessing	Apply Layer Normalization to the input.
linear_filterbanksArchitectures	Return fixed triangular (pre, post) frequency projections as [in, out] kernels.
litert_unavailableConversion	Why this process cannot export LiteRT, one line each; empty when it can.
LiteRTRunnerConversion	Run a single-input, single-output ``.tflite`` model one sample at a time.
LiteRTStreamRunnerConversion	Run a streaming ``.tflite`` model one step at a time, carrying its state as raw tensors.
load_export_recordConversion	Rebuild the Keras model an export record describes, with its weights.
load_modelArchitectures	Loads a Keras model stored either remotely or locally.
load_npyConversion	A ``.npy`` array, memory-mapped, so that a header claiming more data than the file holds is refused
load_pklUtilities	Load pickled file.
load_specConversion	A ``ModelSpec`` from a YAML (``.yaml``/``.yml``) or JSON file.
load_weightsConversion	Load a Keras weights file (``.weights.h5`` or ``.keras``) into ``model``.
mainPackage	Entry point of the ``helia-edge`` command.
make_divisibleArchitectures	Ensure layer has # channels divisble by divisor
MaskedAutoencoderTraining	Masked reconstruction with Keras fit() and independently callable objectives.
MaskedPatchEncoder2DLayers	Project patches with position embeddings and sample masks from a seeded stream.
matches_specUtilities	Test whether data object matches the desired spec.
max_projectionLayers	Magnitude of complex values as the largest of 9 projections of (|re|, |im|), within 0.25% in float32.
maybe_register_serializableUtilities	Register a function or configurable object with Keras serialization.
mbconv_blockLayers	MBConv block w/ expansion and SE
MBConvParamsLayers	MBConv parameters
metaformer_blockArchitectures	Metaformer block
metaformer_layerArchitectures	MetaFormer functional layer
MetaFormerBlockParamsArchitectures	MetaFormer block parameters
MetaFormerParamsArchitectures	MetaFormer parameters
MiniResNetV1ParamsArchitectures	Validated architecture config, independent of inputs and weight assets.
mishLayers	Mish activation layer
mlp_channel_mixerArchitectures	Channel mixer using MLP via 1x1 convolutions
MlperfTinyParamsArchitectures	Select a faithful fixed architecture; initialization and export belong to callers.
mobilenetv1_layerArchitectures	Modified MobileNetV1
MobileNetV1ParamsArchitectures	MobileNetV1 parameters
mobileone_blockArchitectures	MBConv block w/ expansion and SE
mobileone_layerArchitectures	Create MobileOne TF functional model
MobileOneBlockParamsArchitectures	MobileOne block parameters
MobileOneParamsArchitectures	MobileOne parameters
ModePackage	How ``helia-edge export create`` traces the model.
ModelSpecArchitectures	The serializable identity of an architecture.
MultiF1ScoreMetrics	A wrapper around keras.metrics.F1Score to handle multi-dimensional data.
multilabel_confusion_matrix_plotPlotting	Generate multilabel confusion matrix plot via matplotlib/seaborn
NameArgsArchitectures	Name and arguments
no_gradTraining	Disable gradient tracking on Torch; a no-op on other backends.
norm_layerArchitectures	Normalization layer
norm_layerArchitectures	Normalization layer
normalizationLayers	Creates normalization layer based on type
normalizationArchitectures	Normalization layer
Normalization1DPreprocessing	Apply fixed mean/variance normalization to 1D inputs.
Normalization2DPreprocessing	Apply fixed mean/variance normalization to 2D inputs.
NotSupportedTraining	The active Keras backend is not supported by this trainer.
operator_namesConversion	Builtin operator names of every subgraph's operators, in execution order.
parse_factorUtilities	Normalize scalar or paired bounds and validate their range.
patch_embeddingArchitectures	Patch embedding layer using 2D convolution
PatchLayer2DLayers	Extract flattened patches from images shaped (batch, height, width, ch).
plot_history_metricsPlotting	Plot training history metrics returned by model.fit.
pool_token_mixerArchitectures	Token mixer using average pooling
PRDMetrics	Percent RMS difference metric with optional energy normalization.
PrecisionConversion	Numeric format of an exported graph.
px_plot_confusion_matrixPlotting	Generate confusion matrix plot via plotly
random_id_generatorUtilities	Sample with replacement, using optional nonnegative relative weights.
RandomAugmentation1DPipelineAugmentation	Apply repeated random layer choices to an entire batch.
RandomAugmentation2DPipelineAugmentation	The same batchwise composition contract for image transforms.
RandomBackgroundNoises1DAugmentation	Apply random background noises to the input.
RandomChannelAugmentation	Randomly picks a single channel from the input samples.
RandomChoiceAugmentation	Choose one layer for a whole batch; branches must have compatible outputs.
RandomCrop1DAugmentation	Training-only crop; inference leaves the original duration unchanged.
RandomCrop2DAugmentation	Training-only image crop; inference leaves the original extent unchanged.
RandomCutout1DAugmentation	Replace random temporal intervals in each example during training.
RandomCutout2DAugmentation	Replace random rectangular regions in each example during training.
RandomFlip2DAugmentation	Training-only horizontal (width) and vertical (height) image flips.
RandomGaussianNoise1DAugmentation	Apply additive zero-centered Gaussian noise.
RandomNoiseDistortion1DAugmentation	Apply random noise distortion to the 1D input.
RandomSineWaveAugmentation	Adds a sine wave to the input.
readPackage	The tensors of the file at ``path``, read as ``format``.
read_onnxPackage	The initializers of a self-contained ``.onnx`` model, by name.
read_safetensorsPackage	The tensors of a ``.safetensors`` file, by name (little-endian; BF16 is not supported).
read_torchPackage	The tensors of a PyTorch state dict saved with ``torch.save``, loaded with ``weights_only=True``.
ReconstructionTraining	Masked target patches and their corresponding predicted patches.
ReconstructionLossTraining	Reconstruction objective together with the patches used to compute it.
register_keras_serializablesPackage	Register custom objects supported by the selected Keras backend.
regnet_coreArchitectures	RegNet core
regnet_layerArchitectures	Create RegNet TF functional model
RegNetBlockParamArchitectures	RegNet block parameters
RegNetParamsArchitectures	RegNet parameters
reluLayers	ReLU activation layer w/ optional truncation to ReLU6
relu6Layers	Hard ReLU activation layer
reproduceConversion	Rebuild the model a record describes, export it again with the same settings and compare.
ReproductionConversion	`dataclass`
require_backendTraining	Return the active backend, or raise ``NotSupported`` naming ``feature`` and ``supported``.
Rescaling1DPreprocessing	Rescale the input samples.
Rescaling2DPreprocessing	Rescale the input samples.
ReshapePackage	Reshape in C order, as ``numpy.reshape``.
ResidualVectorQuantizerLayers	Residual Vector Quantizer (RVQ) with straight-through estimator.
Resizing1DPreprocessing	Resize signals with bicubic interpolation during training and inference.
Resizing2DPreprocessing	Resize images and aligned data during training and inference.
resnet_layerArchitectures	Generate functional ResNet model.
ResNetBlockParamsArchitectures	ResNet block parameters
ResNetParamsArchitectures	ResNet parameters
resolve_fastenhancerArchitectures	Resolve an official preset with validated overrides into a full record.
resolve_template_pathUtilities	Resolve templated path w/ supplied substitutions.
roc_auc_plotPlotting	Generate ROC plot via matplotlib/seaborn
SamplePreprocessing	`dataclass`
save_pklUtilities	Save python objects into pickle file.
se_blockArchitectures	Squeeze and excite block
se_layerLayers	Squeeze & excite functional layer
set_random_seedUtilities	Set Python, NumPy and selected Keras backend seeds; return the seed.
setup_loggerUtilities	Setup logger with Rich
sigmoidLayers	Sigmoid activation layer
silence_tensorflowUtilities	Silence every unnecessary warning from tensorflow.
SileroBlockStftArchitectures	STFT magnitude of one call (576 samples) from convolutions over 64-sample blocks.
SileroFrameConvArchitectures	An encoder ``Conv1D`` with ReLU over frames held as rows of a (frames, 1, channels) image.
SileroLiveTapsArchitectures	A zero-padded encoder convolution as a dense layer over the kernel taps that see real frames.
SileroVadParamsArchitectures	Silero VAD v6 (16 kHz). The geometry is fixed by the v6.2.2 weights; the options choose how the
SimCLRLossLosses	Implements SimCLR Cosine Similarity loss.
SimCLRTrainerTraining	Creates a SimCLRTrainer.
SnrMetrics	Signal-to-Noise Ratio (SNR) metric where
SourceConversion	A file named by content: its sha256 and, optionally, where to fetch it (never a local path).
SourceFormatPackage	File formats ``import_weights`` reads (see ``helia_edge.importers.readers``).
SourcePinPackage	The one source file a mapping was written for.
SpecAugment2DAugmentation	Mask frequency (height) and time (width), independently per example.
SplitPackage	Take part ``index`` of ``parts`` equal parts along ``axis``.
state_inputLayers	Model input ``state_in_k`` for state pair ``k``.
state_input_nameConversion	Name of the input of state pair ``k``.
state_outputLayers	Name ``x`` as the model output ``state_out_k`` of state pair ``k``.
state_output_nameConversion	Name of the output of state pair ``k``.
state_pairConversion	``("in", k)`` or ``("out", k)`` for a state tensor name, otherwise None.
state_scales_tiedConversion	Whether every integer state pair has one scale and zero point.
StftMagnitudeLayers	Magnitude of a short-time Fourier transform computed as a strided convolution with a stored basis.
stream_calibrationConversion	Calibration for a streaming model: its signal inputs with the states the model produces from them.
StreamingLSTMCellLayers	One LSTM step: ``[x, h, c] -> [h', c']``.
StreamModeUtilities	Select how sample identifiers are scheduled across generator partitions.
SubsampleBlockParamsArchitectures	Subsample block parameters
subsamplerArchitectures	Subsampler block
SumPartsPackage	Add ``parts`` equal parts along ``axis``, for example an ONNX LSTM's input and recurrent biases.
suppress_os_stdioUtilities	Redirects the OS stdout (fd=1) and stderr (fd=2) to /dev/null, then restores them on exit.
swishLayers	Swish activation layer w/ optional hard variant
tcn_block_lgArchitectures	TCN large block
tcn_block_mbArchitectures	TCN mbconv block
tcn_block_smArchitectures	TCN small block
tcn_coreArchitectures	TCN core
tcn_layerArchitectures	TCN functional layer
TcnBlockParamsArchitectures	TCN block parameters
TcnParamsArchitectures	TCN parameters
tensor_recordsConversion	Read the main subgraph's input and output tensors from a ``.tflite`` flatbuffer.
TensorEntryConversion	A model input or output; dynamic dimensions are -1. A state tensor has the index ``pair`` of its
TensorPayloadPreprocessing	Two-level tensor-only schema accepted by portable preprocessing layers.
TensorRecordConversion	`dataclass`
TensorRoleConversion	Role of a model input or output.
threshold_predictionsMetrics	Get prediction indices that are above threshold (confidence level).
tie_state_scalesConversion	Give both tensors of each integer state pair one scale and zero point.
TimePPGParamsArchitectures	TEMPONet-derived PPG heart-rate regressor (Burrello et al., 2022).
to_grainData	Read ``source`` through Grain: shuffle, repeat, transform, batch and prefetch.
to_native_fp16Conversion	Return a graph whose inputs, weights, activations and outputs are FLOAT16.
to_tf_datasetData	Wrap a re-iterable of NumPy batches, such as a Grain dataset, as a ``tf.data.Dataset``.
to_torch_loaderData	Wrap a re-iterable of NumPy batches, such as a Grain dataset, as a Torch ``DataLoader``.
TQDMProgressBarCallbacks	TQDM Progress Bar callback.
TransposePackage	Permute the axes, as ``numpy.transpose``.
TruePRDMetrics	Compatibility wrapper for normalized PRD.
ts_blockArchitectures	Residual block of TSMixer.
tsmixer_layerArchitectures	TsMixer layer
TsMixerBlockParamsArchitectures	TsMixer block parameters
TsMixerParamsArchitectures	TsMixer parameters
undot_layer_namesArchitectures	Rename layers whose names contain ``.`` in a saved Keras model config.
unet_layerArchitectures	Create UNet TF functional model
UNetBlockParamsArchitectures	UNet block parameters
UNetParamsArchitectures	UNet parameters
unext_blockArchitectures	Create UNext block
unext_coreArchitectures	Create UNext TF functional core
unext_layerArchitectures	Create UNext TF functional model
UNextBlockParamsArchitectures	UNext block parameters
UNextParamsArchitectures	UNext parameters
unidentified_installConversion	Why ``record`` does not identify the helia-edge code, and how to install code that does.
uniform_id_generatorUtilities	Yield each ID once per cycle; shuffle mutates the supplied sequence.
use_helia_layer_normalizationArchitectures	Swap Keras ``LayerNormalization`` entries for ``helia_edge.layers.LayerNormalization``.
VectorQuantizerLayers	Vector-quantization bottleneck (VQ-VAE style) with straight-through estimator.
VQAutoencoderTraining	Convenience wrapper around (encoder -> VectorQuantizer -> decoder).
WeightImportConversion	Weights imported from another framework through ``mapping`` (a name in the family's ``MAPPINGS``).
WeightMappingPackage	Every weight of one architecture, from one source format.
WeightRowPackage	One model weight: its source tensors, how they combine, and the transforms to its layout.
weights_digestConversion	``sha256:`` over the model's weights in ``model.weights`` order, independent of the file they came from.
WeightsRecordConversion	The weights, by ``weights_digest``, and their import when they came from another framework.
yblockArchitectures	RegNet Y-Block
zblockArchitectures	RegNet X-Block