search

polarisZhao / PFLD-pytorch

PFLD pytorch Implementation
798 stars 197 forks source link

bad quality #4

Closed iperov closed 5 years ago

iperov commented 5 years ago

python_2019-09-18_22-29-03

iperov commented 5 years ago

python_2019-09-18_22-30-04

iperov commented 5 years ago

python_2019-09-18_22-33-13

iperov commented 5 years ago

@polarisZhao can you please retrain your model with wider variations of scale/rotations/shift,

also add channels shuffling

rnd_state = np.random.RandomState (sample_rnd_seed)
img = np.take (img_bgr, rnd_state.permutation(img_bgr.shape[-1]), axis=-1)

and add random gaussian blur and motion blur

polarisZhao commented 5 years ago

Thanks for your advice, I will add more data augmentation to retain this model.

polarisZhao commented 5 years ago

U can use updeted checkpoint, the landmark will be better, Look forward to your feedback