Registration metrics do not exclude bad frames

[samupicard]

As far as I understand, the registration metrics do not exclude any of the 'bad frames' at the moment: the PCA is performed on an equally subsampled set of frames from the registered movie. This leads to really high metrics on sessions where there were artefacts e.g. the signal dropping suddenly (I've attached a screenshot from the GUI as an example). This approach makes sense for the automatically computed bad frames as you would otherwise underestimate potential registration errors, but surely it would be good to at least take into account the user-defined bad_frames.npy array for these metrics?

It seems that this would be relatively easy fix in this part of the main pipeline code...

suite2p/suite2p/run_s2p.py

Lines 148 to 163 in c88e1ba

	# compute metrics for registration
	if ops.get("do_regmetrics", True) and n_frames >= 1500:
	t0 = time.time()
	# n frames to pick from full movie
	nsamp = min(2000 if n_frames < 5000 or Ly > 700 or Lx > 700 else 5000,
	n_frames)
	inds = np.linspace(0, n_frames - 1, nsamp).astype("int")
	mov = f_reg[inds]
	mov = mov[:, ops["yrange"][0]:ops["yrange"][-1],
	ops["xrange"][0]:ops["xrange"][-1]]
	ops = registration.get_pc_metrics(mov, ops)
	plane_times["registration_metrics"] = time.time() - t0
	print("Registration metrics, %0.2f sec." %
	plane_times["registration_metrics"])
	if ops.get("ops_path"):
	np.save(ops["ops_path"], ops)

[samupicard]

[chriski777]

Thanks for raising this issue, @samupicard! That's a great point you make and we've added a functionality to only remove any user-specified bad_frames before computing the registration metrics. This change should be integrated soon to the main branch but if you'd like to try it out now, feel free to pull the latest changes from the dev branch.

I'll close this issue for now but feel free to reopen if the issue persists.