Error while running XFIQ

[abdu-khamd]

Hey, folks! I encountered the following problem while running the XFIQ.py

  File "XFIQ.py", line 109, in <module>
    run(image_path, model_path, grad_path, T)
  File "XFIQ.py", line 82, in run
    np.save(save_path, save)
  File "<__array_function__ internals>", line 200, in save
  File "/home/shynggys/openvino_2021/lib/python3.8/site-packages/numpy/lib/npyio.py", line 521, in save
    arr = np.asanyarray(arr)
ValueError: setting an array element with a sequence. The requested array has an inhomogeneous shape after 2 dimensions. The detected shape was (2, 3) + inhomogeneous part

I got installed:

Python 3.8
opencv-python=4.6.0.66
Tensorflow=2.7.3
numpy=1.24.0
sklearn=1.2.0
matplotlib=3.6.2
tqdm=4.63.1

[marcohuber]

Dear abdu-Khmad,
only with the error message it is difficult to understand the error. Can you give us more information?
Best,
Marco

[quanqigu]

This is because numpy dosen't support save (string, ndarray, score) direct into numpy.
You need modify the XFIQ.py and explain_quality.py

• XFIQ.py

...
dtype = np.dtype([('path', 'U50'), ('gradients', np.ndarray), ('score', float)])

def run(image_path, model_path, save_path, T):
    ...
    # calculating quality and gradients for each image
    for i in tqdm(images):
        ...
        # add to save
        tmp = (i, grads, score)
        save.append(tmp)

    # make tri-tuple into numpy array
    data_array = np.array(save, dtype=dtype)
        
    # save
    np.save(save_path, data_array)
    
if __name__ == "__main__":
    
    ...
    # Explain Face Image Quality at Pixel-Level
    run(image_path, model_path, grad_path, T)
    loaded_gradients = np.load(grad_path, allow_pickle=True)
    loaded_gradients = loaded_gradients.astype(dtype)
    plot_comparison(loaded_gradients, plot_path, a, b, True)
    

• explain_quality.py

# in plot_comparison, update get names, grads and score
names, grads, score = grad_save['path'], grad_save['gradients'], grad_save['score']

[quanqigu]

I have tested this model with faces captured under surveillance system, most of faces are getting low score. Maybe it is not proper to surveillance system. @pterhoer do you have any suggest?

[pterhoer]

I guess due to the scenario, the general image and face image quality are lower. To still visualize the differences you can adapt the gamma parameter (eq 6).
Best
Philipp