The problem of deep learning record--fast R-CNN training process

Now that the operation has been successful, the problem was not recorded in time, and now can only be written in memory of these problems

Problem 1:matlab command ' MATLAB ' not found. Please add ' matlab ' to your PATH.

Solution: This problem has plagued me for more than a day, obviously the path of MATLAB has been added up, why still appear this problem. Disheartened, can only $fast-rcnn_root\tools\__init__.py 第42-46 line deleted, the problem solved.

However, there are some problems in the training process, do not know whether to delete these lines of code reasons, so you can only change the code back, and then continue to find out why.

Finally found that only need to put the 21 line of $fast-rcnn_root\tools\__init__.py matlab = ' matlab ' to matlab = ' matlab.exe ', the problem of perfect solution

Issue 2:smooth_l1_loss_layer.cpp:19] Check failed:bottom.size () = 4 (3 vs. 4) If weights are used, must specify both and outside weights.

Solution: The $fast-rcnn\data\cache inside the documents are deleted, the train.py inside the Smoothl1loss layer to Smoothl1lossohem

Question 3:typeerror: ' Numpy.float64 ' object cannot be interpreted as a index

Solution: Someone on the internet said pip install statsmodels can solve the problem, but I installed the problem is still there, and netizens said NumPy version is wrong, To install numpy-1.11.0 (my numpy version is 1.12.0) so I downloaded a NUMPY-1.11.3+MKL-CP27-CP27M-WIN_AMD64.WHL, after the installation, the problem solved perfectly.