search

jd-opensource / lapa-dataset

A large-scale dataset for face parsing (AAAI2020)
Other
290 stars 35 forks source link

Some error annotated landmarks in this Datasets #3

Open bestpower opened 4 years ago

bestpower commented 4 years ago

Hi autor, thank you for your sharing this dataset, it is very helpful in our face AI development, but I have found some errors(some annotated face points coordinates is not correct in the corresponding image) in the "LaPa\train\landmarks" when arranging, and list as follows, Please pay attention: 115769026_0.txt 167007578_2.txt 298189567_0.txt 313229012_2.txt 2365836722_8.txt 2384162499_0.txt 3145217556_0.txt 4118951310_0.txt 8982300538_14.txt 11274997965_0.txt 12708537734_6.txt 12829797514_5.txt 13518888075_1.txt HELEN_137346980_1_5.txt HELEN_158249995_1_0.txt HELEN_1437632861_1_4.txt HELEN_2091477588_1_0.txt HELEN_2091682510_1_0.txt HELEN_2454607429_1_1.txt HELEN_2488505181_1_2.txt LFPW_image_train_0671_3.txt LFPW_image_train_0800_0.txt

lexuszhi1990 commented 4 years ago

+1

Yuuuuuuuuuuuuuuuuuummy commented 4 years ago

I meet the same problems, but the incorrect landmark files i found is different with @bestpower . Here is incorrect files list i found:

train:

['HELEN_158249995_1_0', '4079039485_3', '111570982_18', '104951488_10', '1388794206_4', '12829797514_5', '2365836722_8', '5442407218_0', '1134327_0', '2565471402_0', 'HELEN_1437632861_1_4', 'LFPW_image_train_0800_0', 'HELEN_2488505181_1_2', '451124551_0', 'HELEN_2454607429_1_1', '3491434771_3', '3015156301_0', '101777740_2', '7665580390_3', '8132230988_15', '93741462_2', '2512585866_4', '4212777492_2', '12908941174_8', '2064070578_0', '10613204104_2', '2765789940_2', '10412705895_2', '15334793_1']

test:

['2977780949_1', '6620058113_0', '13644546413_2', '398240863_14', '150518966_0', '406855315_3', '161942408_1', '6845488713_5', '294478454_0', '8982300538_14']

just for your reference : )

YuvalNirkin commented 3 years ago

Also it seems that the last two points (on the iris) with indices 104 and 105 are the same as the points with indices 74 and 83.

Anyway thank you for the dataset!