computeCostMatrix method

Tensor computeCostMatrix(
  1. Tensor gtBoxes
)

Implementation

Tensor computeCostMatrix(Tensor gtBoxes) {
  // 1. Validation
  if (shape[1] != 4 || gtBoxes.shape[1] != 4) {
    throw ArgumentError("Both tensors must have 4 columns (x, y, w, h)");
  }

  // 2. Prepare the destination tensor on the GPU
  final int numQueries = shape[0];
  final int numGT = gtBoxes.shape[0];
  final costMatrix = Tensor.fill([numQueries, numGT], 0.0);

  // 3. Call the engine using the internal handles
  // 'gpu' here refers to your CudaEngine singleton/instance
  engine.computeCostMatrix(this.handle, gtBoxes.handle, costMatrix.handle);

  return costMatrix;
}