image quality issue and blurryness at edges of images in semantic segmentation

[shalinikum]

I could not generate the images shown by author in example. I tried different style weight and configurations.But it is still little blurry at edge and lack crispness.

[frostbitten]

What's the full command you're running?

I managed to get the styles to be applied to their respective sections using:

python neural_style.py --content_img 02390.jpg \
                       --content_img_dir ./examples/segmentation \
                       --style_imgs_dir ./examples/segmentation \
                       --style_imgs basquiat.png frida.png \
                       --style_mask \
                       --style_mask_imgs 02390_mask.png 02390_mask_inv.png \
                       --style_imgs_weights 1.0 1.0\
                       --max_size 800 \
                       --max_iterations 600 \
                       --device /gpu:0 \
                       --optimizer adam \
                       --verbose;

with this output:

But this too doesn't quite match the example.

After playing with the parameters (default optimizer, larger style_weight):

python neural_style.py --content_img 02390.jpg \
                       --content_img_dir ./examples/segmentation \
                       --style_imgs_dir ./examples/segmentation \
                       --style_imgs basquiat.png frida.png \
                       --style_mask \
                       --style_mask_imgs 02390_mask.png 02390_mask_inv.png \
                       --style_imgs_weights 1.0 1.0\
                       --max_size 800 \
                       --max_iterations 500 \
                       --device /gpu:0 \
                       --style_weight 1e7 \
                       --verbose;

I can get as close as this:

[cysmith]

I did this project a few years ago so I don't remember. I would try higher style weight and more iterations.