Hi, thanks for making this great work.
The result is very impressive.
I'm trying to reproduce the result of Horse2Zebra and Map2Sat, but I cannot get similar FID values.
Is it normal?
Thanks.
Paper result

Horse2Zebra
The lowest FID is 38.86050602209259.
{'test_155': {'fake_1': 67.28442119755371,
'fake_2': 48.42551658251689,
'fake_3': 42.348939774854806,
'fake_4': 40.1248390131058,
'fake_5': 38.86050602209259},
'test_105': {'fake_1': 76.17878896021992,
'fake_2': 54.2598798947314,
'fake_3': 46.26538334563116,
'fake_4': 42.20495550810186,
'fake_5': 40.45703713549316},
'test_190': {'fake_1': 57.709741354714,
'fake_2': 47.057026217368126,
'fake_3': 45.39016298982065,
'fake_4': 41.18903930884557,
'fake_5': 40.91145373765819},
'test_145': {'fake_1': 66.41517974716504,
'fake_2': 58.00154659998391,
'fake_3': 52.02280410833965,
'fake_4': 47.01486833837396,
'fake_5': 44.17161471652101},
'test_220': {'fake_1': 75.31590592747565,
'fake_2': 56.2749939729235,
'fake_3': 49.92795643094819,
'fake_4': 46.362888518867095,
'fake_5': 45.78734516698335}}
Map2Sat
The lowest FID is 50.47061052521997.
{PosixPath('results/maps_SB/test_375'): {'fake_1': 62.51872202588612,
'fake_2': 53.268906354620356,
'fake_3': 50.47061052521997,
'fake_4': 51.00056969740342,
'fake_5': 54.53886076520848},
PosixPath('results/maps_SB/test_380'): {'fake_1': 63.75456582610627,
'fake_2': 54.06152690467053,
'fake_3': 51.608944166887056,
'fake_4': 52.236401623229256,
'fake_5': 55.30514406652054},
PosixPath('results/maps_SB/test_315'): {'fake_1': 61.31653824103762,
'fake_2': 53.38316725431366,
'fake_3': 51.89277393992245,
'fake_4': 54.22393842199415,
'fake_5': 59.4284568657705},
PosixPath('results/maps_SB/test_400'): {'fake_1': 64.5273366808837,
'fake_2': 54.21571474249083,
'fake_3': 51.989683097455384,
'fake_4': 53.5245091661194,
'fake_5': 56.79759882286055},
PosixPath('results/maps_SB/test_370'): {'fake_1': 62.95248898270029,
'fake_2': 54.47948824315364,
'fake_3': 52.69446956424861,
'fake_4': 55.165577782972775,
'fake_5': 60.10190538034851}}
Commands
Horse2Zebra
Train Command
python train.py --dataroot ./datasets/horse2zebra --name h2z_SB --mode sb --lambda_SB 1.0 --lambda_NCE 1.0 --gpu_ids 0 --display_env h2z_SB
Test command
#!/usr/bin/env bash
set -e
for epoch in {5..400..5}
do
python test.py \
--dataroot ./datasets/horse2zebra \
--name h2z_SB_2 \
--checkpoints_dir ./checkpoints \
--mode sb \
--eval \
--phase test \
--num_test 1000 \
--epoch "$epoch" \
--gpu_ids 0
done
Map2Sat
Train Command
python train.py --dataroot ./datasets/maps --name maps_SB --mode sb --lambda_SB 1.0 --lambda_NCE 1.0 --gpu_ids 1 --direction BtoA --display_env maps_SB
Test command
#!/usr/bin/env bash
set -e
for epoch in {5..400..5}
do
python test.py \
--dataroot ./datasets/maps \
--name maps_SB \
--checkpoints_dir ./checkpoints \
--mode sb \
--eval \
--phase test \
--direction BtoA \
--num_test 2000 \
--epoch "$epoch" \
--gpu_ids 1
done
Hi, thanks for making this great work.
The result is very impressive.
I'm trying to reproduce the result of Horse2Zebra and Map2Sat, but I cannot get similar FID values.
Is it normal?
Thanks.
Paper result
Horse2Zebra
The lowest FID is
38.86050602209259.Map2Sat
The lowest FID is
50.47061052521997.Commands
Horse2Zebra
Train Command
python train.py --dataroot ./datasets/horse2zebra --name h2z_SB --mode sb --lambda_SB 1.0 --lambda_NCE 1.0 --gpu_ids 0 --display_env h2z_SBTest command
Map2Sat
Train Command
python train.py --dataroot ./datasets/maps --name maps_SB --mode sb --lambda_SB 1.0 --lambda_NCE 1.0 --gpu_ids 1 --direction BtoA --display_env maps_SBTest command