fix(show1by1): --output_folder is parsed but images always (+1 more)

[andrewwhitecdw]

Small fixes in visual_tools/show1by1.py:

fix: --output_folder is parsed but images always written to hardcoded ./visual_data/rainbow

Fix: Replace:

save_path = './visual_data/rainbow'
if not os.path.isdir(save_path):
    os.mkdir(save_path)

with:

save_path = opts.output_folder

fix: --a2b option is ignored; only the a-side generator/ID models are used

Fix: Replace:

state_dict_gen = torch.load(opts.checkpoint_gen)
trainer.gen_a.load_state_dict(state_dict_gen['a'], strict=False)
trainer.gen_b = trainer.gen_a

state_dict_id = torch.load(opts.checkpoint_id)
trainer.id_a.load_state_dict(state_dict_id['a'])
trainer.id_b = trainer.id_a

trainer.cuda()
trainer.eval()
encode = trainer.gen_a.encode # encode function
style_encode = trainer.gen_a.encode # encode function
id_encode = trainer.id_a # encode function
decode = trainer.gen_a.decode # decode function

with:

state_dict_gen = torch.load(opts.checkpoint_gen)
trainer.gen_a.load_state_dict(state_dict_gen['a'], strict=False)
trainer.gen_b.load_state_dict(state_dict_gen['b'], strict=False)

state_dict_id = torch.load(opts.checkpoint_id)
trainer.id_a.load_state_dict(state_dict_id['a'])
trainer.id_b.load_state_dict(state_dict_id['b'])

trainer.cuda()
trainer.eval()
if opts.a2b == 1:
    encode = trainer.gen_a.encode # encode function
    style_encode = trainer.gen_a.encode # encode function
    id_encode = trainer.id_a # encode function
    decode = trainer.gen_a.decode # decode function
else:
    encode = trainer.gen_b.encode # encode function
    style_encode = trainer.gen_b.encode # encode function
    id_encode = trainer.id_b # encode function
    decode = trainer.gen_b.decode # decode function

Files changed

• visual_tools/show1by1.py

[andrewwhitecdw]

Closing this sweep-generated PR: PR has 2 commits; sweep requires exactly one commit per PR. It does not meet the sweep requirements (single signed-off commit).