Images location
Didier0 opened this issue · comments
@ankush-me
Hi, where are the 5 images (hiking, indian+musicians, sandwich, sea, village) that Synthtext script uses saved, from where the script load this images? Is there some file or? Ty
Please check here.
Yea but how to run .h5 file? With matlab or?? I don't understand that .h5 file
I am not sure what you mean by 'run[ning]' that file. That h5 file is used by the generation script to get information about the background images.
Okay but how did you put images into this file? Yes I checked here "https://github.com/ankush-me/SynthText#generating-samples" but I can't figure out. Four days I can't understand how to add my own background images. Can you just tell me what do I need to do to add images into .h5 file or how to open this file or I don't know? Where I put my own images??
I am sorry this has been difficult for you.
As mentioned in the readme (under the heading 'Adding New Images'), for adding new images, you need to obtain their segmentations and depth values. Once you have them you need to modify the script (here: Lines 102-113) to read that information in. It is not required to dump the new images in an h5 file -- it is just given as a container for the 5 sample images. You can store the image, segmentation, depth data for each new image as you desire as long as there are loaded correctly in the generation script.
Okay I get it, so if I understand I need get depth value from predict_depth.m and segmentation value from run_ucm.m and floodFill.py? Then these scripts generate .h5 file and that .h5 file script uses to read info of images?
And then this values I put in the script?
Yes that is correct in principle, but again, generating a new h5 file is not required -- you can just load the three things directly. Also note that predict_depth.m and run_ucm.m are now quite dated --- it will be difficult to make them run (as they depend on external codebases (that too in MATLAB!) which may not be maintained anymore). It will be easier to use more modern alternatives to both of these.
Also 8K images with their segmentation and depth values are provided --- check the readme.