TensorFloat32Views class
Cached Float32List views of an Interpreter's input and output tensors.
Capture once after Interpreter.allocateTensors and reuse the views on
every inference. The views alias the underlying tensor native memory,
so writing to inputs stages data directly into the input tensor and
reading from outputs reads the latest values after
Interpreter.invoke (or runForMultipleInputs).
This avoids recreating Float32List wrapper objects every inference
and avoids repeated getInputTensor / getOutputTensor lookups on
the hot path.
Validity:
- Interpreter.allocateTensors must have been called before capture.
- Views become stale after any Interpreter.resizeInputTensor or
subsequent
allocateTensors. Recapture in that case. - Only meaningful for tensors whose element type is float32. For
quantized tensors, interpreting bytes as Float32 will produce
garbage; use the raw
Tensor.databytes instead.
Example:
interp.allocateTensors();
final views = TensorFloat32Views.capture(interp);
// Hot path:
views.inputs[0].setAll(0, preparedNHWC);
interp.invoke();
final scores = views.outputs[1]; // aliases tensor native memory
Constructors
- TensorFloat32Views.capture(Interpreter interp)
-
Captures Float32List views of every input and output tensor of
interp.factory
Properties
- hashCode → int
-
The hash code for this object.
no setterinherited
-
inputs
→ List<
Float32List> -
One Float32List view per input tensor, in declaration order.
final
-
outputs
→ List<
Float32List> -
One Float32List view per output tensor, in declaration order.
final
- runtimeType → Type
-
A representation of the runtime type of the object.
no setterinherited
Methods
-
noSuchMethod(
Invocation invocation) → dynamic -
Invoked when a nonexistent method or property is accessed.
inherited
-
toString(
) → String -
A string representation of this object.
inherited
Operators
-
operator ==(
Object other) → bool -
The equality operator.
inherited