search

TreB1eN / InsightFace_Pytorch

Pytorch0.4.1 codes for InsightFace
MIT License
1.72k stars 418 forks source link

about Normalize([0.5, 0.5, 0.5], [0.5, 0.5, 0.5] #135

Open kidwhh opened 4 years ago

kidwhh commented 4 years ago

I saw that Normalize([0.5, 0.5, 0.5], [0.5, 0.5, 0.5] was used for the input data when the train was made.Face_verify.py is also used, why is Normalize not used when evaluating agedb_30, cfp_fp, LFW. Is there something special about the data set, or something else?

haolinlyou commented 4 years ago

I saw that Normalize([0.5, 0.5, 0.5], [0.5, 0.5, 0.5] was used for the input data when the train was made.Face_verify.py is also used, why is Normalize not used when evaluating agedb_30, cfp_fp, LFW. Is there something special about the data set, or something else?

when you run prepare.py to create test_dataset (agedb_30, cfp_fp, LFW), it will invoke func load_bin from data_pipe.py. Then you can find it in data_pipe.py line 61 data[i, ...] = transform(img) The func transform play a role of normalize.