Extend the solver to support joint estimation of a shared PSF across multiple blurred measurements:
y_i = h * x_i,\quad i=1..B
This is necessary for “self-calibration” and significantly improves kernel identifiability.
Acceptance Criteria
• Modify BlindDeconvolver.initialize_from_measurement to allow B > 1
• Kernel parameter k should remain shape (1,1,Kh,Kw) and get broadcast over batch
• Modify forward_model() to support B>1 via grouped conv
• Update testbench to optionally stack multiple measurements with the same kernel
Extend the solver to support joint estimation of a shared PSF across multiple blurred measurements:
y_i = h * x_i,\quad i=1..B
This is necessary for “self-calibration” and significantly improves kernel identifiability.
Acceptance Criteria
• Modify BlindDeconvolver.initialize_from_measurement to allow B > 1
• Kernel parameter k should remain shape (1,1,Kh,Kw) and get broadcast over batch
• Modify forward_model() to support B>1 via grouped conv
• Update testbench to optionally stack multiple measurements with the same kernel