possible memory leak in lsqr

In motion estimation global displacement lsqr robust I encountered OOM issues, similar to those encountered with the lsmr. Maybe a leak. I also fixed with gc.collect() but for some reason I had issues with the import. This is my code block:

        # TODO: this is already soft_weights
        import gc
        xrange = trange if progress_bar else range
        for i in xrange(lsqr_robust_n_iter):
            p = lsqr(A[idx].multiply(W[idx]), V[idx] * W[idx][:, 0])[0]
            idx = np.nonzero(np.abs(zscore(A @ p - V)) <= robust_regression_sigma)
            import gc
            gc.collect()
            # print('temporary memory cleared')
        displacement = p
        gc.collect()
        print('temporary memory cleared')

I never bothered to figure out which of these two did the job. Just an FYI

Locate the lsqr implementation used by motion estimation global displacement and reproduce the OOM with the reported robust-iteration loop. Compare memory behavior with and without gc.collect(); done means identifying whether a leak exists and documenting or testing a fix, but the report names no source file or test.