The web demo has the following defects:
- The parameter mean_file should be mean in the classifier initialization in app.py
- The mean is treated as a string when it should be an array in pycaffe.py (should add mean = np.load(mean) above the line if mode == 'elementwise':
- Even after making the code run, the actual result doesn't work, that is, no matter the query image, the result is classified as a "window screen." (I believe the problem lies in the python/caffe code)