IndexError when runing data_preparation.py

[li-fang]

Hi! I tried to train the network on my own datasets. When I was preparing the landmarks by running the script "python data_preparation.py --img_folder <folder_to_training_images>", I got the following error message for one of the images:

Traceback (most recent call last):
File "data_preparation.py", line 48, in
data_prepare([os.path.join(opt.data_root, folder) for folder in opt.img_folder], opt.mode)
File "data_preparation.py", line 28, in data_prepare
detect_68p(img_folder, lm_sess, input_op, output_op) # detect landmarks for images
File "D:\Code\Python\Deep3DFaceRecon_pytorch\util\detect_lm68.py", line 105, in detect_68p
draw_landmarks(img, landmark, os.path.join(vis_path, name))
File "D:\Code\Python\Deep3DFaceRecon_pytorch\util\detect_lm68.py", line 29, in draw_landmarks
:] = np.array([0, 0, 255])
IndexError: index 432 is out of bounds for axis 0 with size 432

I changed corresponding index "img.shape[0]-landmark[i, 1]+j" to "img.shape[0]-1-landmark[i, 1]+j". And the above error didn't occur for the next try. I'm not sure if that's what you really meant.

BTW, thanks for sharing your work!

[YuDeng]

Hi, thanks for pointing out this problem. "img.shape[0]-1-landmark[i, 1]+j" should be a better form to avoid out of bound issue if a detected landmark is quite close to image boundary. Alternatively you can clip the result within [0, image.shape-1].

This should have no influence to the training process because draw_landmarks function is only used for visualization purpose.

[li-fang]

Thank you for your clarification!