Open mlo0352 opened 3 years ago
I've discovered that you can point evaluate.py to a directory containing a style.py output (the four checkpoint files you mentioned) and the script will be able to use that to render your input image.
For example if you put the four checkpoint files into the checkpoints/style01 directory you can then invoke it like this:
python evaluate.py --checkpoint checkpoint/style01/ --in-path content.jpg --out-path styledimage.jpg
I imagine it'll only work if there's one set of checkpoint files in there.
But I don't know how to combine those four files into a single checkpoint file like @lengstrom did. There's no documentation for it.
Did you do this But I don't know how to combine those four files into a single checkpoint file like @lengstrom did. There's no documentation for it. ?????????
I ran style.py and it terminated without error. I see some files that were created in my checkpoint directory:
checkpoint fns.ckpt.data-00000-of-00001 fns.ckpt.index fns.ckpt.meta
I cannot use any of these files to run with evaluate.py, however. Am I missing something? Is there something I can do to create the .cpkt file so I can run it?Here's how I ran style.py for reference:
python style.py --style style_image/footprints_test_1000.jpg --checkpoint-dir checkpoint_dir/ --test test_image/30073094_10160273796265076_666433414413834817_o.jpeg --test-dir test_dir/ --content-weight 1.5e1 --checkpoint-iterations 1000 --batch-size 20
Here's the error I get when trying to point to any of the files in the checkpoint/ directory:
Traceback (most recent call last): File "evaluate.py", line 212, in <module> main() File "evaluate.py", line 209, in main batch_size=opts.batch_size) File "evaluate.py", line 102, in ffwd saver.restore(sess, checkpoint_dir) File "/root/anaconda3/envs/tf-gpu/lib/python3.7/site-packages/tensorflow_core/python/training/saver.py", line 1290, in restore {self.saver_def.filename_tensor_name: save_path}) File "/root/anaconda3/envs/tf-gpu/lib/python3.7/site-packages/tensorflow_core/python/client/session.py", line 960, in run run_metadata_ptr) File "/root/anaconda3/envs/tf-gpu/lib/python3.7/site-packages/tensorflow_core/python/client/session.py", line 1183, in _run feed_dict_tensor, options, run_metadata) File "/root/anaconda3/envs/tf-gpu/lib/python3.7/site-packages/tensorflow_core/python/client/session.py", line 1361, in _do_run run_metadata) File "/root/anaconda3/envs/tf-gpu/lib/python3.7/site-packages/tensorflow_core/python/client/session.py", line 1386, in _do_call raise type(e)(node_def, op, message) tensorflow.python.framework.errors_impl.DataLossError: Unable to open table file checkpoint_dir/checkpoint: Data loss: not an sstable (bad magic number): perhaps your file is in a different file format and you need to use a different restore operator?