matlab.engine

[SEAFLYFISH]

大佬您好！我是在conda创建的虚拟环境运行的，在基本配置完毕运行huston denoise时他告诉我
ERROR: Could not find a version that satisfies the requirement matlab.engine (from versions: none)
ERROR: No matching distribution found for matlab.engine
意思是需要从matlab创建虚拟环境吗，还是说是我之前安装包的时候出错了

[LiPang]

可能是matlab和python的版本没有对应

[Ares960826]

It is recommended to replace the relevant code segment with the following code:

# ------------- new import -------------
from scipy.linalg import qr                                         # NEW
# --------------------------------------

... ... ... ...

    # --------- REPLACED: MATLAB ‑> SciPy pivoted-QR implementation -----
    if not opt['no_rrqr']:
        # Instead of calling MATLAB's sRRQR_rank we use SciPy's pivoted QR
        # to pick the Rr * K most informative spectral bands.
        #
        # E has shape (Ch, Rr*K)
        # We work on its transpose so that QR decomposes the bands (columns).
        E_np = E[0].cpu().numpy().T                                    # (Rr*K, Ch)
        _, _, piv = qr(E_np, pivoting=True, mode='economic')           # piv is a permutation array
        selected = np.sort(piv[:Rr * K])                               # keep first Rr*K pivots, sort asc
        param['Band'] = torch.tensor(selected, dtype=torch.int64,      # move to torch / GPU
                                     device=device)
    else:
        param['Band'] = th.tensor([Ch * i // (K * Rr + 1)
                                   for i in range(1, K * Rr + 1)],
                                   dtype=torch.int64,
                                   device=device)
    print(param['Band'])
    # ----------------- end of replacement section ---------------------

Original version:

    if not opt['no_rrqr']:
        import matlab.engine
        eng = matlab.engine.start_matlab()
        eng.cd(r'matlab')
        res = eng.sRRQR_rank(E[0].cpu().numpy().T, 1.2, 3, nargout=3)
        param['Band'] = th.Tensor(np.sort(list(res[-1][0][:3]))).type(th.int).to(device)-1

    else:
        param['Band'] = th.Tensor([Ch * i // (K * Rr + 1) for i in range(1, K * Rr + 1)]).type(th.int).to(device)
    print(param['Band'])

from Mingze Yang
Ph.D. Candidate
Video and Hyperspectral Imaging Lab
Center of Signal and Image Processing (CeSIP)
University of Strathclyde, Glasgow, UK